Conceptual Atomism and the Computational Theory of Mind

Advances in Consciousness Research (AiCR)

Provides a forum for scholars from different scientific disciplines and fields of
knowledge who study consciousness in its multifaceted aspects. Thus the Series
includes (but is not limited to) the various areas of cognitive science, including
cognitive psychology, brain science, philosophy and linguistics. The orientation
of the series is toward developing new interdisciplinary and integrative
approaches for the investigation, description and theory of consciousness, as well
as the practical consequences of this research for the individual in society.
From 1999 the Series consists of two subseries that cover the most important
types of contributions to consciousness studies:
Series A: Theory and Method. Contributions to the development of theory and
method in the study of consciousness;

Editor
Maxim I. Stamenov
Bulgarian Academy of Sciences

Editorial Board
David J. Chalmers Stephen L. Macknik
Australian National University Barrow Neurological Institute, Phoenix, AZ,
Gordon G. Globus USA
University of California at Irvine George Mandler
Ray Jackendoff University of California at San Diego
Brandeis University Susana Martinez-Conde
Christof Koch Barrow Neurological Institute, Phoenix, AZ,
California Institute of Technology USA

Stephen M. Kosslyn John R. Searle

Harvard University University of California at Berkeley

Earl Mac Cormac Petra Stoerig

Duke University Universität Düsseldorf

Volume 69
Conceptual Atomism and the Computational Theory of Mind. A defense of
content-internalism and semantic externalism
John-Michael Kuczynski
Conceptual Atomism and
the Computational Theory of Mind
A defense of content-internalism
and semantic externalism

John-Michael Kuczynski
Virginia Commonwealth University

John Benjamins Publishing Company

Amsterdam / Philadelphia
 TM The paper used in this publication meets the minimum requirements of
8

American National Standard for Information Sciences – Permanence of

Paper for Printed Library Materials, ansi z39.48-1984.

Library of Congress Cataloging-in-Publication Data

Kuczynski, John-Michael.
Conceptual atomism and the computational theory of mind : a defense of content-
internalism and semantic externalism / John-Michael Kuczynski.
p. cm. -- (Advances in consciousness research, issn 1381-589X ; v. 69)
Includes bibliographical references and index.
1. Philosophy of mind. 2. Knowledge, Theory of. 3. Cognitive science. 4. Atomism.
5. Internalism (Theory of knowledge) 6. Externalism (Philosophy of mind) I.
Title.
BD418.3.K83   2007
128'.2--dc22 2007008592
ISBN 978 90 272 5205 0 (hb : alk. paper)

© 2007 – John Benjamins B.V.

No part of this book may be reproduced in any form, by print, photoprint, microfilm, or any
other means, without written permission from the publisher.
John Benjamins Publishing Co. · P.O. Box 36224 · 1020 me Amsterdam · The Netherlands
John Benjamins North America · P.O. Box 27519 · Philadelphia pa 19118-0519 · usa
 Table of contents

Introduction ix

Part I
A defense of content-internalism and of a descriptivist theory of concepts

chapter 1
Basic concepts 3

chapter 2
The predicative nature of sense-perception 25

chapter 3
Uniquely individuating descriptions 41

chapter 4
Some semantic consequences or our analysis: Tokens versus types,
semantics versus pre-semantics 69

chapter 5
Modality, intensionality, and a posteriori necessity 97

chapter 6
Cognitive maps and causal connections: Why the causal story is an
important part of the descriptive story 109

chapter 7
Conception as knowledge of series of interlocking existence-claims 123

chapter 8
The problem of de re senses 129

chapter 9
Publicity problems and the nature of linguistic communication 153
 Conceptual Atomism and the Computational Theory of Mind

chapter 10
Content-externalism and self-knowledge 163

chapter 11
Why one’s mental content is fixed by one’s epistemic situation 181

chapter 12
Jackson and Pettit on program-causality and content-externalism 189

Part II
Fodor, conceptual atomism, and computationalism

chapter 13
Content-externalism and conceptual atomism 217

chapter 14
The concept of a symbol 261

chapter 15
Event-causation and the root-problem with the Computational
Theory of Mind 285

chapter 16
Fodor’s first argument for conceptual atomism 293

chapter 17
Fodor’s second argument for conceptual atomism 315

chapter 18
Fodor’s third argument for conceptual atomism 327

chapter 19
Some arguments for the Symbolic Conception of Thought 397

chapter 20
A positive argument against the Symbolic Conception of Thought 409

chapter 21
Another argument against the Symbolic Conception of Thought:
The concept of non-conceptual content 419
 Table of contents 

chapter 22
Propositional structure and the ineliminability of non-conceptual content 429

chapter 23
Conceptual content and the structure of the proposition 443

chapter 24
Peacocke on concept-possession 467

chapter 25
Semantics versus psychology 483

Conclusion 507

Bibliography 509

Index 517
 Introduction

A doctrine known as “content-externalism” is widely taken to provide a correct analy-

sis of mental content. A doctrine known as the Computational Theory of Mind (CTM)
is widely taken to provide a correct analysis of mental operations. These doctrines sup-
port each other.
The present work is divided into two parts. In Part I (Chapters 1–12) it is argued
that content-externalism is false, but that a related doctrine known as “semantic exter-
nalism” is correct. An alternative to content-externalism is proposed.
In Part II (Chapters 13–25) it is argued that CTM is false. Special attention is paid
to the concept of syntactic form. It is seen that CTM involves a misunderstanding of
this concept, and that this misunderstanding is embedded in misunderstandings that
are reinforced by content-externalism.
part I

A defense of content-internalism and

of a descriptivist theory of concepts
 chapter 1

Basic concepts

Content-externalism

Let W and W* be two different possible worlds satisfying the following conditions. In
W, Max sees rock R at time t. In W*, Twin-Max sees rock R* at the same time. R is
numerically distinct from R*. But apart from that, there are no differences between
W at t and W*at t. (It follows that Max and Twin-Max are numerically identical.) So
even though R and R* are numerically distinct, they nonetheless look the same, and
have the same mass, shape, and so on. And even though Max is looking at one rock,
whereas Twin-Max is looking at some other rock, the physiological disturbances that
result in the one case are qualitatively identical with those that result in the other.
Of course, Max and Twin-Max do not have exactly the same relational properties.
For example, Max’s visual experience at time t is caused by R, not R*, and Twin-Max’s
corresponding experience is caused by R*, not R. Nonetheless, Max and Twin-Max are
obviously extremely similar. So long as we consider the regions of space-time occupied
by their bodies, there is nothing that distinguishes Max from Twin-Max. If there is an
electron-jump in the region occupied by Max, there is a qualitatively identical elec-
tron-jump in the region occupied by Twin-Max.
According to a widely held doctrine, Max’s perceptions and subsequent thoughts
don’t have the same contents as Twin-Max’s corresponding perceptions and thoughts.
Here is the reasoning behind this claim:
Max’s visual perception is veridical exactly if R has certain properties, it being ir-
relevant whether R* has those properties. And Twin-Max’s visual perception is
veridical exactly if R* has certain properties, it being irrelevant whether R has
those properties. Similarly when Max thinks that rock is lovely, his thought is cor-
rect exactly if R is lovely, it being irrelevant whether R* is lovely. And when Twin-
Max thinks that rock is lovely, his thought is correct exactly if R* is lovely, it being
irrelevant whether R is lovely. Therefore Max’s R-perceptions and subsequent
thoughts don’t have the same truth-conditions, or therefore the same contents, as
Twin-Max’s R*-perception and subsequent-thoughts.
Leaving aside facts about the origins of their conditions, Max and Twin-Max are
qualitatively identical. So it is solely in virtue of the fact that they are embedded in
different environments that they have perceptions and thoughts with different con-
tents. Thus, in at least some cases, a mental state’s content is fixed, at least in part, by
facts about the origins of that state. Two brain-states can be qualitatively identical,
leaving aside facts about their causal origins, and yet have different contents. Thus,
 Conceptual Atomism and the Computational Theory of Mind

what a person (or creature) is thinking is constitutively (and not just causally) de-
pendent on facts about his environment, the same being true of the information en-
coded in his sense-perceptions.1

The doctrine just described is known as “content externalism.” The present work is con-
cerned with content-externalism (among other things). We will argue that it is false.
Content-externalism must at all costs be distinguished from another doctrine that
we will refer to as “semantic externalism.”

Semantic externalism

A continuation of our story about Max and Twin-Max will help us define the term
“semantic externalism.” Max later tells people about R, referring to it as “Rocko.” Be-
cause Max is an eminent person, “Rocko” becomes part of the English language. So
people who have never seen R refer to it as “Rocko”, and they say things like: “the rock
in your backyard is smaller than Rocko” and “Rocko is supposedly the most beautiful
rock in the world.” Thus, in W, “Rocko” refers to R, i.e. it becomes a semantic rule of
English that “Rocko” refers to R (just as, in actuality, it is a semantic rule of English that
“Socrates” refers to Socrates). Everything that we just said about Max is true mutatis
mutandis of Twin-Max. Thus, in W*, “Rocko” refers to R*.
Here is what semantic externalism says about this situation. In W, an utterance of
“Rocko weighs over five lbs” has for its literal meaning the singular proposition R
weighs over five lbs. In general, for any predicate phi, an utterance in W of FRocko has
phiG has for its literal meaning the singular proposition R has phi. For exactly similar
reasons, an utterance of FRocko has phiG in W* has for its literal meaning the singular
proposition R* has phi. Thus, what is literally meant by an utterance in W of FRocko
has phiG neither entails, nor is entailed by, what is literally meant by a homonymous
utterance in W*. (Salmon 1986, Kaplan 1989, Soames 1989, 2001, 2005).
Content-externalism and semantic-externalism are distinct doctrines.2 The first is a
doctrine about what people think. The second is a doctrine about what our words mean.
As we will see, a failure to distinguish between these two doctrines has led to much con-
fusion in contemporary thought. We will find that semantic-externalism is unqualifiedly
true, even though content-externalism is unqualifiedly false.
There are many well-known apparent problems with semantic-externalism. For ex-
ample, if semantic-externalism is right, then “Hesperus=Hesperus” has precisely the
same literal meaning as “Hesperus=Phosphorous.” But surely, it will be said, these sen-

1. See Burge (1979, 1982), Kaplan (1989b: 322-326), McGinn (1986), Salmon (1986), and Da-
vidson (2004: 77-86).
2. Blackburn (1984: 328-333) makes this point.
 Chapter 1.  Basic concepts 

tences must have different meanings, given that one of them is trivial whereas the other
is non-trivial.
We will find that this problem, and others like it, vanish when we take a few basic
facts into account – in particular, the fact that one must always exploit background
semantic and non-semantic knowledge to compute the meanings of the sentence-to-
kens that one encounters.

Some apparent problems with content-externalism

As we have already seen, there is much to be said for content-externalism. It is a datum

that, notwithstanding their similarities, Max and Twin-Max are seeing, and subse-
quently thinking about, different rocks. So it seems a datum that the contents of Max’s
mental states are, to that extent, different from the content’s of Twin-Max’s mental
states. Finally, there is no way to explain these differences except in terms of the fact
that Max’s mental states don’t have the same origins as Twin-Max’s. These differences
have no intra-cranial basis, and must therefore be sought in what is extra-cranial, i.e.
in what is external.
But there is a problem. So long as we confine ourselves to the regions occupied by
their bodies, there is nothing that distinguishes Max from Twin-Max. It immediately
follows that they are physically identical. And unless we take the dubious measure of
saying that their minds engulf regions of space larger than those encompassed by their
bodies3, it also follows that they are psychologically identical.4 So to the extent that it
must be understood along content-externalist lines, mental content is without any psy-
chological significance: what a person is thinks, believes, and doubts is irrelevant to
how that person is psychologically. But it would seem to be a truism that two people
differ psychologically if they differ in respect of what they believe, think, or doubt. As
we will see, some authors reject this supposed truism on the grounds that it conflicts
with content-externalism. But it is prima facie to the discredit of that doctrine that it
has such a revisionist consequence.
Before moving on, we should consider a possible externalist rejoinder to the argu-
ment just given:
You say that Max’s mind is confined to his cranium, and that to think otherwise is
“dubious.” But in saying this, you are simply begging the question. The essence of
content-externalism is that some facts about Max’s mind are fixed by extra-cra-
nial facts. So if that doctrine is correct, it is not absurd to say that someone’s mind
literally encompasses some remote region of space-time; and it is thus not absurd

3. This position is taken by McGinn (1986: 20-30).

4. Stich (1978: 590-591) powerfully explains why it is not an option, even for the content-ex-
ternalist, to say that Max and Twin-Max are psychologically different.
 Conceptual Atomism and the Computational Theory of Mind

to say that Max and Twin-Max have qualitatively different minds. So I don’t see
that you’ve made any headway here against content-externalism.5

It is easy to show that this response has consequences too revisionist to accept. Because
light travels at a finite speed, one often sees stars that went out of existence long ago.
Given this, suppose that, instead of being nearby rocks, R and R* are qualitatively iden-
tical but numerically distinct stars that went out of existence 100 million years ago. In
that case, the objector’s response amounts to this: at least some portion of Max’s mind
was around 100 million years ago. But that is straightforwardly false. There were no
human minds back then.
This point can be generalized. One sees states of affairs, not objects. You don’t just
see Smith. Rather, you see a dated situation that has Smith as a component. Because
light travels at a finite speed, anything that you see has since ceased to exist, and any-
thing that is contemporaneous with your visual perception is, at most, a successor or
continuation of the state of affairs that you are seeing. Since content-externalism makes
the thing you are seeing into a veritable constituent of the content of your perception, it
follows that the objector’s response drains all of your perceptions of any content. Given
that our perceptions obviously do have content, we may conclude that Max and Twin-
Max are psychologically and physically identical, contrary to what the objector says.

Some apparent problems with content-externalism (continued)

Suppose that Max does, whereas Twin-Max does not, believe that there is a nearby
ATM machine that is spitting out hundred dollar bills. There will be an immediate dif-
ference between Max and Twin-Max in regards to what they feel, want, and do. Max
will, whereas Twin-Max will not, experience a crisis of conscience (“should I take
money that is not mine?”). Max will, whereas Twin-Max will not, repress feelings of
guilt that result from committing theft. Max does, whereas Twin-Max does not, de-
velop a feeling of self-loathing, which is defensively cloaked under a veneer of exces-
sive self-confidence. And so on.
Here, we have a situation where Max’s mental content clearly differs from Twin-
Max’s. That initial difference quickly ramifies into others, and Max and Twin-Max for-
ever cease to have parallel lives.

5. McGinn holds that content-externalism is correct and that, consequently, one’s mind is
spread out all over the cosmos. See McGinn (1986: 20-30). McGinn then argues that this conse-
quence of content-externalism is innocuous since the mind is not a “substance.”
It is extremely unclear what it means to say that the mind is not a “substance.” But what is clear
is that there are representational mental events. Once this is granted, it follows (as we will now see)
that, if content-externalism is right, parts of your mind existed millions of years ago. It follows that
content-externalism is false, given that your mind came into existence only a few decades ago.
 Chapter 1.  Basic concepts 

To see why this is a problem for the content-externalist, let us return to the previ-
ous scenario – the one where the only differences between Max and Twin-Max coin-
cide with, or derive from, the fact that the one person saw R while the other saw R*. In
that scenario, the alleged difference in content doesn’t lead to any lack of parallelism.
Even if we concede to the content-externalism that Max and Twin-Max have different
psychological contents, those contents are perfectly parallel.
For reasons of a purely methodological nature, that perfect parallelism is suspi-
cious. We are reminded of Poincare’s famous thought-experiment. Let W and W* be
two different universes. (These needn’t coincide with the two homonymously named
universes previously discussed.) At time t, everything in W undergoes a tenfold in-
crease in size. Apart from that, W and W* are qualitatively identical.
Is there any difference between W and W*? Poincaré famously said “no”, and his
answer was vindicated by later developments in both physics and logic. Because the
changes that occurred in W were perfectly uniform, the relational properties of events in
the one world remain identical with those of events in the other world. Thus, I am twice
as tall as my niece in W iff I am twice as tall as she is in W*. If my pants are too tight in
W, then they are too tight by exactly the same amount in W*. It is obvious that these
points generalize without limit. There is thus no conceivable way of differentiating be-
tween W and W*, and the statement “I am in W, as opposed to W*” is without any ex-
planatory or experimental significance. So we are guilty of making a distinction without
a corresponding difference if we say that those worlds differ. It follows that the notion of
absolute size, and therewith of absolute position (and velocity and mass…), are empty.6
The differences between Max and Twin-Max are like those between the worlds in
Poincare’s thought-experiment. Genuine differences destroy parallelism and disrupt
relational facts. But the supposed differences between Max and Twin-Max are com-
pletely non-disruptive. Those two individuals remain in perfect lock-step (Fodor 1987,
Jackson and Pettit 2004: 45–46).
A content-externalist might respond by saying:
You are begging the question. Max and Twin-Max don’t remain in lock-step. Max
has dreams about R, whereas Twin-Max has dreams about R*. Max wants to build
a miniature replica of R, whereas Twin-Max wants to build a miniature replica of
R*. And these differences do ramify into others. Given that both Max and Twin-
Max believe that there are no square circles, Max comes to believe either R weighs
over five lbs or there are square circles whereas Twin-Max comes to believe either

6. Matters are actually not quite as simple as this. Given only the just-described similarities
between W and W*, it is still possible that they are dynamically different, even though kinema-
tically they are the same. See Schlick (1919) and Reichenbach (1958). The thought-experiment
has to be developed further to generate the result that Poincaré wants. But these subtleties are
obviously not relevant here.
 Conceptual Atomism and the Computational Theory of Mind

R* weighs over five lbs or there are square circles. So contrary to what you said, we
don’t have one difference: we have a whole pattern of differences.7

The content-externalist wishes to show that, because Max’s perception results from R
as opposed to R*, his mental content is qualitatively different from Twin-Max’s. All of
the differences cited by the objector presuppose that this is the case. So the differences
cited by the objector in defense of the content-externalist’s position presuppose the
truth of that position, and thus don’t provide any independent support for it.

Content-internalism versus content-externalism

A theory is a way of modeling data. In philosophy, these data consist of intuitions. For
example, we all have the intuition that killing infants is evil; and any viable ethical
theory must either validate that intuition or it must show that, relative to intuitions of
an even more fundamental kind, it can be explained away.
It is a datum that some of Max’s thoughts are made true by facts about R, and that
Twin-Max’s corresponding thoughts are made true by facts about R*. It is also a datum
that there are certain apparent similarities between Max’s R-thoughts and Twin-Max’s
R*-thoughts. Content-externalism provides one model of such data. Content-internal-
ism rejects that model, and must therefore propose an alternative to it.8 (By “content-
internalism” I mean the doctrine that content-externalism is false.) This work will pro-
pose a content-internalist model of the data in question.
These days most philosophers believe that content-externalism is correct. In other
words, most philosophers believe that, because of the differences between W and W*,
Max and Twin-Max differ in respect of the identity of the information encoded in their
sense-perceptions and subsequent thoughts.
For the reasons given a moment ago, it seems very reasonable to believe that there is
a kind of representational content that Max and Twin-Max have in common, and that
this common content is not affected by the differences between their two worlds. Follow-
ing convention, we will refer to that kind of content (supposing that it exists) as “narrow
content” (Loar 1985, Jackson and Pettit 2004c). So supposing that individuals x and y are
qualitatively identical, leaving aside facts about the origins of their conditions, x and y
cannot differ in respect of the narrow contents of their perceptions and thoughts.
Of course, all content-internalists believe that there is such a thing as narrow-con-
tent, and so do some content-externalists. Such content-externalists are forced to ad-

7. This is John McDowell’s position. See McDowell’s commentary on Chapter 6 of Evans’ Va-
rieties of Reference (Evans 1982: 203-204).
8. It is an open question in the philosophy of science whether there can be two different but
equivalent ways of modeling the same data.
 Chapter 1.  Basic concepts 

vocate a “two-tiered” or “two-dimensional” conception of mental content. We will re-

fer to this position as “soft content-externalism.”9
A number of eminent content-externalists deny that there is such a thing as nar-
row-content. Of these, the most prominent are probably Tyler Burge, Gareth Evans,
and John McDowell. We will refer to this position as “hard-line content-externalism.”
In this work I will argue that both forms of content-externalism are quite false. I
will also argue that semantic externalism is entirely correct. We will see that many of
the grounds for content-externalism vanish as soon as that doctrine is sharply distin-
guished from semantic externalism.
Jerry Fodor (1998) is a hard-line externalist who has proposed a very definite view
as to the nature of conception, and he has presented a number of powerful arguments
for that view. Indeed, given an acceptance of hard-line content-externalism, Fodor’s
analysis of conception follows, as we will see in a moment.
The essence of Fodor’s position is this. For individual x to have a concept of object
y is for some of x’s states to have been caused, in a certain way, by states of affairs in-
volving y. (In other words, there is some causal relation R such that x has a concept of
y just in case some state of affairs involving y R-caused x to undergo changes of a cer-
tain kind.) Fodor’s view can be understood in terms of the following scenario. I see
Fred. This involves some state of affairs involving Fred causing me to have certain psy-
chological states (various sensory experiences and attendant thought-processes). In
their turn, those states lead to my being able to think about Fred, i.e. they lead to my
subsequently having a concept of Fred. Here we have a case where one thing’s having a
concept of some other thing involves the second thing’s having some kind of effect on
the first thing. Since it is an obvious fact that there exist situations like the one just
described, nobody denies that conception may involve a causal connection between
subject and object. But Fodor goes a step further, saying that any instance of concep-
tion not only involves, but is identical with, an instance of the kind of causal relation
just discussed. My having a concept of Socrates not only involves, but is identical with,
my being on the receiving end of a causal chain that begins with some state of affairs
involving Socrates.10 Fodor attempts to show that my concepts of theoretical entities
(e.g. muons) and also of abstract entities (e.g. the property of justice) can be explained
along similar lines.11
Of course, this conception of conception is not an easy one to accept. But Fodor
provides independent corroboration for it. More specifically, he successfully shows
that each of three widely held and independently plausible doctrines implicitly presup-
poses the truth of his analysis (Fodor 1975, 1987, 1994, 1998). Those doctrines are the
Computational Theory of Mind (CTM), the Symbolic Conception of Thought (SCT),

9. Loar (1985) is a soft content-externalist. So are Jackson and Pettit (2004b). Chalmers’ (1996)
argument for dualism involves a version of soft content-externalism.
10. Fodor (1998).
11. Fodor (1990).
 Conceptual Atomism and the Computational Theory of Mind

and content-externalism. SCT is the view that we think in symbols and thus in some
kind of language. CTM is the view that all thinking is computing. In due course, we
will say more completely what SCT and CTM are.
Given that each of these doctrines presupposes the truth of Fodor’s analysis, it fol-
lows that there is at least as much support for the latter as there is for the best sup-
ported of those three doctrines. Further, there is, as we will see, considerable indepen-
dent support for each of these doctrines. So even though Fodor’s analysis might
initially seem counterintuitive, there is much to be said in support of it.

Conceptual atomism and content-externalism

As I will be using the word “concept”, x has a concept of y iff x can have thoughts about
y. Because you are able to think about Socrates, you have a concept of the man.
If correct, Fodor’s analysis warrants the acceptance of a doctrine that we will refer to
as “conceptual atomism.” Consider your concept of Socrates. It would seem, on the face
of it, that you can have that concept only because you already have various others. Sup-
posing that X is a creature that has no conception of space or time or corporeality, could
X have a concept of Socrates? Most of us would say “no.” It seems absurd to suppose that
a creature so totally lacking in cognitive wherewithal could possibly have a concept of
Socrates. More generally, it seems absurd to suppose that, for anything x, an otherwise
conceptless creature could have a concept of x. It seems, therefore, that one’s concept of
Socrates (or of anything else) is not an “atom”, and is necessarily embedded in a network
of other concepts. Some kind of conceptual non-atomism seems obligatory.12 One of the
objectives of this work is to show that conceptual non-atomism is in fact correct.
But if Fodor’s analysis is right, conceptual atomism is not only permissible, but de
rigueur. According to that analysis, my having a concept of Socrates consists solely in my
having a certain causal connection with Socrates. Whatever causal connection I have with
Socrates could presumably be had by an otherwise conceptless being, e.g. a photographic
plate. So Fodor’s analysis seems to allow that such a being could have a concept of Socrates.
In general, it seems that if we accept a strictly causal analysis of conception, such as that
advocated by Fodor, then we are committed to conceptual atomism.
Fodor recognizes this and, true to his own premises, he accepts conceptual atom-
ism. Fodor says that an otherwise conceptless creature could have a concept of Socrates
– or of my desk or of justice or of the color green. Because he is keenly aware that many

12. Sellars (1963) made a strong case for non-atomism. Non-atomism is now generally, but not
universally, accepted. Bonjour (1985) discusses why non-atomist must be accepted. Fodor
(1998) argues that it must be rejected. In Kuczynski (2003, 2004), I provide non-Sellarsian argu-
ments for non-atomism. Elsewhere (Kuczynski 2007c), I have argued that a version of Sellars’
argument was given by Berkeley in A Theory of Vision and, in fact, that Berkeley’s version is su-
perior to Sellars’.
 Chapter 1.  Basic concepts 

will disagree with this atomism, he attempts to find independent corroboration for it.
The merits of these attempts will be evaluated in Chapters 16–18.
Fodor’s atomism is easily shown to be at least conditionally correct. (Here we will out-
line an argument that we will give in full in Chapter 13.) Suppose that content-externalism
is correct. In that case, two people can be qualitatively identical but for the fact that one of
them has one concept where the other has some different concept. According to the con-
tent-externalist, Max has a concept of R instead of a concept of R*, even though, apart from
that, Max and Twin-Max are qualitatively identical. If all concepts were constitutively
linked with other concepts, then it wouldn’t be possible for two people to differ with re-
spect to just one concept. So to the extent that mental content is to be understood along
content-externalist lines, concepts come one at a time, and not in clusters.
A brief digression may help. Suppose that members of species X don’t have hearts.
In that case, it won’t be a biological possibility that members of that species are other-
wise just like human beings. Members of that species must differ systemically from
people – otherwise they would be in the same health-situation as a person without a
heart, and would thus immediately go out of existence.
According to the non-atomist, what we just said about hearts is true of concepts. If
Smith has the concept of a conversion-reaction whereas Jones does not, then Smith
must have other concepts that Jones does not. Jones’ acquiring the concept of a conver-
sion-reaction involves his acquiring other concepts – e.g. the concepts of repression, of
a punitive super-ego, and of unconscious mental activity. The non-atomist says that any
acquisition of a single concept involves an acquisition of multiple concepts; and this is
equivalent to the view, stated a moment ago, that conceptual differences between peo-
ple cannot be confined to a single concept. (We will develop this argument in Chapter
13.) As we’ve seen, content-externalism presupposes the falsity of that view.
So given content-externalism, Fodor’s atomism follows, as does his strictly causal
conception of conception, as we will now see. According to the content-externalist,
Max has a concept of R, this being the one concept that Max has that Twin-Max lacks.
Max is causally connected to R, this being the one thing that distinguishes Max from
Twin-Max. Of course, exactly the same points mutatis mutandis hold of Twin-Max.
The conceptual differences between Max and Twin-Max supervene wholly on the dif-
ferences in their causal liaisons. The fact that Max does, whereas Twin-Max does not,
have a concept of R is wholly a function of the fact that Max is, whereas Twin-Max is
not, causally connected to R in a certain way. So to the extent that the content-exter-
nalist account of mental content is the right one, it follows that differences in concepts
are to be understood in strictly causal terms, and therefore that conception itself is so
to be understood. So given an acceptance of content-externalism, Fodor’s atomism
and his strictly causal conception of conception both follow.13

13. This general line of thought lurks beneath the arguments found in Fodor (1990). But it is
not explicitly stated there. Nor, to my knowledge, is it explicitly stated anywhere in the externa-
list (or, for the matter, the internalist) literature.
 Conceptual Atomism and the Computational Theory of Mind

Thus, Fodor is simply being true to his own premises in accepting these two doc-
trines. Most other content-externalists have not gone so far as to say that conception is
to be understood in purely causal terms or that a strictly atomistic view of concepts is
correct. But so far as this is the case, the systems advocated by those other content-
externalists are not consistent ones.

The relation of CTM to conceptual-atomism

But Fodor’s primary reason for accepting conceptual atomism is not that content-ex-
ternalism demands it. His primary reason is that he accepts CTM and, as Fodor him-
self cogently argues, the truth of CTM depends on that of atomism. (Fodor (1968,
1975) accepted CTM before content-externalism even came into existence.) In Chap-
ters 13–22 we will argue that CTM is not viable and that the arguments for it are falla-
cious. Let us briefly outline what we will say.
For the sake of argument, suppose that CTM is right. In other words, suppose that
all thinking is “computing.” A question immediately arises: What exactly does the word
“compute” mean in this context?
Right now, I could compute a sum – I could add 134 to 397. But that kind of com-
putation obviously presupposes various cognitive faculties, and thus cannot underlie
all cognition.
Fodor is aware of this, and he defines the word “computation” in a way that, he
believes, does not render CTM viciously circular. A computation, he says, is a “formal
operation on symbols” (Fodor 1987: 19). It is an operation that is causally driven by the
forms, not the meanings, of the symbols involved. A machine can respond to the form
of the expression “23+15” in such a way that it produces the right output. No thinking
is needed to perform a purely formal operation. Given that computing is a purely for-
mal operation, it follows that there is no vicious circularity in supposing that thinking
is computing (Fodor 1981: 13–17, Fodor 1987: 18–20).
An immediate consequence of CTM – indeed, the very essence of it – is that think-
ing is not content-driven (Fodor 1987: 19). Of course, such a position initially strikes
us as very odd. I learn that Bill fell off of a tall building, and I immediately thereafter
form the belief that Bill is now injured or deceased. Surely my believing the proposi-
tion Bill fell off of a tall building led, or at least might have led, to my believing the
proposition Bill is now injured or deceased. Thus, at least some sequences of thoughts
are content-driven. So supposing that we think in symbols, it is very hard to believe
that the semantic properties of those symbols are causally inert.
But Fodor (1987, 1990) has produced an argument that, if cogent, shows why we
mustn’t put too much stock in this particular criticism of CTM. The purely formal prop-
erties of symbols can be coordinated with their representational properties in such a way
that all of the relevant facts are easily explained – e.g. the fact that, if I believe that Bill
has seven cars, then I am likely to form the belief that Bill has a prime number of cars.
 Chapter 1.  Basic concepts 

True – this account does strip the mental per se of causal efficacy. But theories are to be
evaluated in terms of their degree of concordance with the relevant data, not in terms of
their concordance with our a priori conceptions. In this case, the relevant datum is that
certain kinds of thoughts are likely to follow other kinds of thoughts. It is not a datum
that semantics is what is causally responsible for these patterns.

The concept of syntax: an introduction

There are a number of points to make in response to the argument just outlined. I will
now only outline these points, since they will be discussed at length later on.
Supposing that a computation is understood to be a “form-driven” operation, we
must ask: what is meant by the term “form”? Sometimes Fodor says that what he has in
mind is syntactic form. So Fodor’s position becomes: a computation is a syntax-driven
operation. Is this claim feasible?
Before we can answer this question, we must note that the term “syntax” is am-
biguous: what linguists mean by it isn’t what mathematical logicians mean by it. Let us
now see if either disambiguation of the term “syntax” validates Fodor’s position. Ac-
cording to a linguist, the sentences:
(*) “John punched Fred”
and
(**) “Larry kissed Mary”
have similar (or identical) syntactical structures. What does this mean? It means that they
have similar derivation-trees. To say that they have similar derivation-trees is to say that
the (complex) semantic rule that assigns meaning to the one sentence is structurally sim-
ilar to the semantic rule that assigns meaning to the other. Those sentences are syntacti-
cally similar because of how they are assigned their respective meanings by the semantic
rules of English. In general, the syntax of a sentence lies not in what it means, but in how
it means what it means. Syntax is meaning-how. (Semantics is meaning-what.) The con-
cept of syntax is therefore not a meaning-innocent one.
So supposing, as Fodor does, that thinking is an operation on expressions, syntax-
driven operations are meaning-driven. They are driven by a sensitivity not to what the
operands mean, but to how they mean what they mean. Sensitivity to meaning-how, no
less than sensitivity to meaning-what, presupposes a heavy cognitive arsenal. It is therefore
not tenable to suppose that syntax-driven operations could be cognitively foundational.
Let us give a different version of the argument just outlined. The term “wet” is not
interchangeable with the expression “the property of wetness.” After all, “one shouldn’t
drive on wet roads” is meaningful, whereas “one shouldn’t drive on the property of
wetness roads” is not. If you know only that the term “wet” denotes or expresses a cer-
 Conceptual Atomism and the Computational Theory of Mind

tain property, and do not know the syntactic properties of that expression, then you are
ignorant of a crucial aspect of its meaning. Syntax is thus an aspect of meaning.
More precisely, syntax is combinatorial semantics. One knows the syntax of a
complex expression when one knows the combinatorial semantic properties of its con-
stituents. Equivalently, one knows the syntax of an expression when one knows how its
meaning is derived from those of its constituents – when, in other words, one knows
its derivation tree. Thus, an operation is syntax-driven iff it is driven by a sensitivity to
combinatorial semantics. Since such an operation can be carried out only where there
is already an awareness of meaning, it is not an option to hold that such operations
form the foundation of our cognitive lives.
The linguist’s disambiguation of the term “syntax” thus fails to validate Fodor’s
position. Let us now see if Fodor’s position is validated by the mathematical logician’s
disambiguation of that term.
First of all, how are the two disambiguations different? Most linguists would say that
(1) “Bob is tall or Bob is not tall”
has the same syntax as
(2) “Bob is tall and Bob is not tall.”
The linguist sees “or” as being in the same syntactic category as “and.” From his stand-
point, (1) and (2) are no more syntactically different than “Bob is tall” and “Bob is
short.” But the mathematical logician says that, because (2) has an “or” in the place
where (1) has an “and”, those sentences are in different syntactic categories. What is the
basis of this (apparent) disagreement?
When a mathematical logician says that a sentence S, belonging to language L, is
“syntactically true”, he means that it is a theorem or consequence of the semantic rules
of L that S is true.14 And when a mathematical logician says that two sentences S and
S* have the same syntax, he means that they are interchangeable from a proof-theo-
retic standpoint. Equivalently, S1 and S2 have the same “syntax” iff….S1…is syntacti-
cally true iff the same is true of…S2….(So, for example, if there is some sentence S3
such that S1→S3 is a semantic theorem and S2→S3 is not, then S1 is syntactically different
from S2.) It is a theorem of the rules of English semantics that (1) is true, but not that
(2) is true. (Indeed, it is a theorem of those rules that (2) is false.) And that is why, from
a mathematical logician’s standpoint, S1 and S2 are syntactically different.
Even though the mathematical logician’s disambiguation of the term “syntax” is dif-
ferent from the linguist’s, the former poses the same problems for Fodor’s analysis as the
latter. On both disambiguations, a sentence’s syntax lies in how it means what it means.
Of course, the linguist and the logician don’t have quite the same thing in mind
when they think about how a sentence is assigned the meaning that it has. The latter is

14. We will defend this claim in Chapter 13. It is anticipated by Carnap (1937, 1956) and Gödel
(1953).
 Chapter 1.  Basic concepts 

concerned with non-psychological notions such as provability; the former is con-

cerned with psychological notions such as comprehension and learnability. But in both
cases, a sentence’s syntax lies in a relationship that holds between the semantic rules of
the relevant language and the sentence’s meaning. So given either disambiguation of
the term “syntax”, syntax-driven operations are meaning-driven. This immediately en-
tails that Fodor’s position is a non-starter

Indexicality and the impossibility of a syntactic characterization of logical form

There is an entirely different reason why CTM is in trouble if “form” is identified with
syntax. For CTM to work, it must be possible to coordinate syntax with semantics. But
there is no disambiguation of the term “syntax” that makes such a coordination pos-
sible. A linguist would say that:
(^) “Smith is ill because Smith doesn’t smoke.”
has the same syntax as
(^^) “Smith is ill and Smith doesn’t smoke.”
Given this sense of the word “syntax”, if some syntax-driven entity accepts each of
“Smith is ill” and “Smith doesn’t smoke”, it is just as likely to transition to (^) as it is to
transition to (^^). But this clearly doesn’t correspond to the transitions that a rational
person would make. So the linguist’s definition of “syntax” fails to validate the idea that
thinking consists in syntax-driven operations.
According to the mathematical logician, (^) and (^^) do have different syntactic
structures. Indeed, from the mathematical logician’s viewpoint, the “syntax” of a sen-
tence is by definition identical with its logical form. So it would seem that this disam-
biguation of the term “syntax” would validate Fodor’s analysis.
But this is not so. Relative to this reading of the term “syntax”, syntax tracks logical
form only if the language in question satisfies conditions that would make that lan-
guage useless in the way of mediating rational thought. In that sense of the word “syn-
tax”, an utterance of “that [accompanied by an ostension of Smith] is identical with
that [accompanied by an ostension of Jones]” has the same syntax as an utterance of
“that [accompanied by an ostension of Smith] is identical with that [accompanied by
another ostension of Smith].” So a purely syntax-driven entity wouldn’t be capable of
registering the profound logical differences between these two utterances. Since it is
obvious that human beings can do so, it follows that, in this sense of the word “syntax”,
we aren’t syntax-driven entities (Kuczynski 2006b).
What we just said about “that” is true of “here”, “now”, “this”, “over there”, and any
other context-sensitive expression. Syntax can be coordinated with logical form only
where there is no context-sensitivity. So supposing that we think in a language whose
 Conceptual Atomism and the Computational Theory of Mind

syntactic structures are coordinated with their logical forms, it follows that those sen-
tences have no context-sensitive component and, consequently, that one isn’t capable of
thinking things like it is now 3:00 p.m. or that guy is Jones. Since we obviously do have
such thoughts, our thoughts are not mediated by a language of the kind just described.

Syntax as morphology

Sometimes when Fodor talks about sentential “form”, he is referring to the property of
having a certain shape. He is referring to form in a purely geometrical sense (Fodor
1987: 18–20).
But this disambiguation of the word “form” doesn’t validate the contention that
thought is syntax-driven or, therefore, that it is meaningfully described as
computational. Syntax is not morphology. Syntax doesn’t supervene on morphology.
Syntax is no more capable than semantics of being coordinated with morphology. (We
will argue for these claims in Chapter 14.) So in direct opposition to what Fodor holds,
thought is not syntax-driven, and is therefore not computational, to the extent that it
is morphology-driven.
To be sure, there is a long tradition of identifying syntax with morphology. But
that tradition is mistaken, and it involves erroneous views as to the meanings of ex-
pressions such as “syntax”, “algorithm”, “mechanical procedure”, and “formal truth.” It
might be appropriate to outline what we will say in defense of these claims. Given two
sentence-tokens that have the same syntax, there is no limit to how different they can
be in respect of morphology. And given two sentence-tokens that have the same mor-
phology, there is no limit to how different they can be in respect of syntax. So syntax
isn’t morphology, and it isn’t realized by morphology. It is true that, within narrow
contexts, a “morphological characterization” of syntax is possible, making it appear as
though morphology-driven processes are syntax-driven. But this appearance is mis-
leading. To say that there is a “morphological characterization” syntax is to say that
there is a function that assigns syntactic structures to symbol-tokens on the basis of
their shapes. Given this, it is clear why, even in a context where a morphological char-
acterization of syntax is possible, no process can possibly be syntax-driven in virtue of
its being morphology-driven. Suppose that, in possible world W, all round wheels are
red and all square wheels are green. In W, a vehicle with red wheels would ceteris pari-
bus function better than a vehicle with green wheels; and we could say that, in W, there
is a “chromatic characterization” of what it is to be a good wheel. But the red wheels
aren’t good because they are red – they are good because they are round. Similarly even
if all and only those expressions having a certain shape have a certain syntax, they don’t
have that shape because they have that syntax. A corollary is that, even in contexts
where morphology tracks syntax, it doesn’t do so because morphology is (or realizes)
syntax. Thus, morphology-driven processes are no more syntax-driven than, in W,
instances of successful transportation are color-driven.
 Chapter 1.  Basic concepts 

The idea that formal (morphology-driven) processes are also formal (syntax-driv-
en) involves a failure to see that the term “form” is ambiguous between two totally dif-
ferent meanings. The terms “mechanical procedure”, “algorithm”, and “formal truth” are
comparably ambiguous and are used in a comparably equivocal manner by CTM. Once
these equivocations are exposed, that doctrine loses its intuitive appeal and the argu-
ments for it crumble.

The root-problem with CTM and content-externalism

But there is a problem with CTM, and also with content-externalism, more fundamen-
tal than any thus far mentioned. It is states of affairs, not objects, that make things
happen. It is not the rock that breaks the window – it is the rock’s colliding with the
window. In light of this point, suppose that brain-state B realizes a condition of believ-
ing that Socrates was bald. There is no doubt that B has causal properties: it has a cer-
tain shape and a certain mass; it consists of particles with certain charges; and so forth.
But , for the reasons given earlier, B’s encoding the proposition that Socrates was bald
is causally inert, at least so far as content is to be understood along externalist lines,
and B’s having that content thus becomes as explanatorily idle as its having the prop-
erty of being a thing x such that x is either a circle or not a circle (Jackson and Pettit
2004: 46–48, Kuczynski 2006c).
It follows that content-externalism threatens the very existence of mental content.
Where the occupants of the spatiotemporal are concerned, existence and causal potency are
indistinguishable. Fodor (1968) himself makes this point very clearly,15 and so does Jaegwon
Kim (1993: 348). A state of affairs that has no causal powers is no state of affairs at all. Since
content-externalism strips mental content of causal powers, it denies its existence.
John McDowell argues that content-externalism is compatible with the causal ef-
ficacy of the mental.16 But McDowell’s argument is a straightforward non-starter, as we
will see in a moment.
Fodor sees clearly that, if content-externalism is right, no brain-state has causal
powers in virtue of its representational properties; i.e. he sees that content-externalism
strips content of causal efficacy. (That is why, in his view, it is the “syntactic structures”
of brain-states, not their semantic (representational) properties, that do all the causal
work.) But Fodor seems not to see that this jeopardizes the very existence of represen-
tational, and therefore mental, properties.
Like McDowell, Frank Jackson and Philip Pettit hold that content-externalism is
consistent with the causal potency of the mental (Jackson and Pettit 2004–2004c). But
their argument is very different from McDowell’s, and has considerably more merit.

15. Of course, spatiotemporal points and lines, and so forth, lack causal powers. But they are
not occupants of space-time. Rather, they are sets of space-time coordinates.
16. See McDowell’s commentary in Evans (1982: 203-204).
 Conceptual Atomism and the Computational Theory of Mind

Their argument is based on a distinction between what they call “program-causality”

and “efficient-causality.” They argue that content-externalism allows representational
states of affairs (e.g. a brain-state’s being a belief that Plato was wise) to be program-
causes, even though it doesn’t allow them to be efficient-causes.
I believe that the Jackson-Pettit analysis of causality is accurate and that the dis-
tinction between program-causality and efficient-causality is of the highest impor-
tance. But their worthy analysis of causality does not show that content-externalism is
compatible with the presumption representational states of affairs have any causal
properties: it no more allows mental states to be program-causes than it allows them to
be efficient causes. This will be shown in Chapter 12.

Thee key distinctions

In this work, it will be argued that, so far as content-externalism appears plausible, that
is because we are failing to make the following three distinctions:
(1) The distinction between what is literally meant by our expressions and the
information that is imparted to us in the process of computing their literal
meanings.
(2) The distinction between content and truth-maker.
(3) The distinction between perceptual and meta-perceptual data (between what
is encoded in our perceptions, taken by themselves, and what we infer from
our perceptions).
Since much of Part I will be spent delineating (1)-(3), I will now only give an outline of
what they mean. For the sake of argument, suppose that, as Kaplan (1989) holds, de-
monstrative expressions are “directly referential.” In that case, the proposition literally
meant by a token of:
(A) “that overbearing and generally unsavory man standing over there next to
that atrocious painting is a professor of anthropology”
is simply:
(B) x is a professor of anthropology,
for some value of x. But obviously what is communicated by a token of (A) will not be
so threadbare. What is communicated is (inter alia) that somebody is overbearing and
generally unsavory and is also standing next to some atrocious painting. Thus, what is
communicated by a token of (A) will involve concepts like painting, overbearing, unsa-
vory that are quite absent from (B). How is this to be reconciled with Kaplan’s thesis
that (B) is the literal meaning of (A)?
 Chapter 1.  Basic concepts 

There is no problem. We need only take into account the fact that people typically
understand the expressions that they encounter on the basis of their knowledge of the
semantic rules that assign them meaning. The semantic rule for the demonstrative ex-
pression “that overbearing and generally unsavory man standing over there next to
that atrocious painting” is:
(C) For any context C, and any predicate phi, if somebody x in C is uniquely a
salient overbearing and generally unsavory man standing next to a uniquely
salient painting, then a token in C of Fthat overbearing and generally unsa-
vory man standing over there next to that atrocious painting has phiG means:
x has phi.
The rule for (A) specifically is:
(D) For any context C, if somebody x in C is uniquely a salient overbearing and
generally unsavory man standing next to a uniquely salient painting, then a
token in C of “that overbearing and generally unsavory man standing over
there next to that atrocious painting is a professor of anthropology” means: x
is a professor of anthropology.
When one hears a token of (A), one has to work through (D). Anyone who knows (D)
will know that an utterance of (A) will not be true unless, in the context of utterance,
somebody x is a uniquely salient overbearing and generally unsavory man standing
next to a uniquely salient atrocious painting and, moreover, x is a professor of anthro-
pology. So even though what a token of (A) literally means is (B), and thus doesn’t in-
volve the concepts of being unsavory or overbearing, what such a token communicates
to someone does concern these very concepts. Because of the information that one
must work through in order to assign the right proposition to a token of (A), what such
a token communicates is very different from what it literally means.
I will refer to information conveyed in this way as “pre-semantic implicature.” We
will find that what we just said about tokens of (A) is true, to some degree or other, of
all sentences. In connection with this, we will find that pre-semantic implicatures are
exponentially more powerful than the post-semantic implicatures studied by Grice.
Let us now briefly say what is meant by (2) and (3). Consider the sentence:
(S) “Somebody invented bifocals.”
This sentence is true because Benjamin Franklin invented bifocals. So Franklin’s in-
venting bifocals is the truth-maker of (S). But the meaning of (S) obviously isn’t: Frank-
lin invented bifocals. After all, even though it is false, “somebody invented bifocals, but
Franklin did not do so” is not self-contradictory.17 What we just said about sentences is
true of perceptions and thoughts. When I look at an ice-cube, what makes my percep-
tion veridical is that there are various H2O-molecules in a certain configuration in a

17. I am borrowing an argument given by Russell (1919: 165).

 Conceptual Atomism and the Computational Theory of Mind

certain region of space-time. So the truth-maker of my perception is given by some

proposition having the form: in region R, there are various molecules consisting of hy-
drogen and oxygen atoms, and those molecules are interrelated in such and such a man-
ner. But what my perception tells me isn’t given by such a proposition. So the content of
my perception doesn’t involve the concepts molecule, oxygen, or hydrogen.
Distinction (3) is easily understood in terms of these points. Given my sense-
perception, along with my knowledge of chemistry, I can infer that there are various
molecules of a certain kind in front of me. But, as we just saw, the concepts molecule,
oxygen, and so on, are no part of the content of my perception. We mustn’t confuse
what, given one’s background knowledge, one can infer from a given perception with
what that perception itself tells one; we mustn’t confuse perceptual data with post-
perceptual meta-data.
In his landmark work the Varieties of Reference, Gareth Evans put forth deep and
original analyses of conception, semantic content, and the relationship between the
two. I believe that, by combining Evans’ insights with (1)-(3), we are able to produce
models of both semantic and cognitive content that satisfy two important requirements.
First, those models enable us to hold onto all of the pre-theoretic intuitions that are
prima facie inconsistent with consent-externalism. For example, those models enable
us to hold onto our intuition that, in at least some cases, our beliefs about our own
minds are characterized by a kind of certainty and incorrigibility that could not possi-
bly characterize our beliefs about the external world. (As we will see, content-external-
ism cannot accommodate this datum.) Second, those models enable us to account for
the important fact, stressed (but misunderstood) by content-externalism, that our con-
cepts of spatiotemporal objects have an ineliminable causal component.
Evans’ notion of a “cognitive map” is going to play a crucial role in our analysis of
conception. We will also find important applications for Evans’ distinction between
“conceptual” and “non-conceptual” content. Evans’ idea of a de re sense (Evans 1985:
Chapter 10) – which came to have an important place in McDowell’s work (McDowell
1998: Chapters 10–13) – will also be discussed at length. We will find that it is an inco-
herent synthesis of a number of deep insights.

McDowell on mental causality

We will end this chapter by substantiating a point made earlier. As previously noted,
John McDowell holds that content-externalism is consistent with the fact that the
mental is causally efficacious. McDowell’s position can be understood in terms of the
following story. Let w* be a world that is just like ours except that w* contains XYZ,
instead of H2O. Let Smith be a person in our world, and let Smith* be Smith’s doppel-
ganger in w*. In our world, Smith’s perception of a certain glass of H2O causes him to
reach for that glass. In w*, Smith*’s perception of a certain glass of XYZ causes him to
reach for that glass.
 Chapter 1.  Basic concepts 

Here is what McDowell says about this situation. Smith’s perception has a different
content from Smith*’s. There is some glass x such that the content of Smith’s perception
does, whereas the content of Smith*’s perception does not, have x has a constituent. And
there is some glass y such that the content of Smith*’s perception does, whereas the con-
tent of Smith’s perception does not, have y has a constituent. Further, there is some natu-
ral kind N – namely, H2O – such that the content of Smith’s perception does, whereas the
content of Smith*’s perception does not, have N has a constituent. Finally, there is some
natural kind N* – namely, XYZ – such that the content of Smith*’s perception does,
whereas the content of Smith*’s perception does not, have N* has a constituent.
(Before we continue, we should note that we’ve already seen some reason to believe
that the contentions just stated are wrong, as they involve a conception of perceptual
content that is rendered incoherent by its failure to recognize some key distinctions –
in particular, the distinction between data and meta-data, between literal and cogni-
tive meaning, and between content and truth-maker. But let us leave that aside for
now, and continue with our exposition of McDowell’s analysis.)
According to McDowell, it is obvious why content-externalism is consistent with
the fact that the mental has causal powers. In virtue of having glass x, as opposed to
glass y, for its content, Smith’s perception has causal powers not had by Smith*’s per-
ception. Smith’s perception does, whereas Smith*’s perception does not, cause some-
body to reach for glass x. And in virtue of having glass y, as opposed to glass x, for its
content, Smith*’s perception has causal powers not had by Smith’s perception. Smith*’s
perception does, whereas Smith’s perception does not, cause somebody to reach for
glass y. Similar remarks explain why, in virtue of having H2O, as opposed to XYZ, for
part of their contents, Smith’s post-perceptual mental states (e.g. his intention not to
make loud slurping noises when drinking the contents of the glass in question) have
causal powers not had by Smith*’s post-perceptual mental states.
But contrary to what McDowell says, the scenario just described is consistent with
the view that the causal properties of Smith’s perception and post-perceptual states are
identical with those of Smith*’s corresponding states. Magnet A is attracted to magnet
B. Magnet B happens to be green. But ceteris paribus if B were red or purple or orange,
A would still be attracted to it. It is true that, because B is green, A is behaving in a way
that it would not behave if B were some other color. (A is drawn towards a green mag-
net, instead of a red one.) But it doesn’t follow that B’s color has anything to do with A’s
behavior. A’s behavior wouldn’t ceteris paribus be different in a world where B were
some other color. In other words, A’s behavior wouldn’t be relevantly different. This
shows that B’s being green has no effect on what A does, notwithstanding that, because
B is green, A is doing something (namely moving towards a green magnet) that it
wouldn’t be doing if B were red.
Similarly, given only that Smith’s behavior is caused by H2O, as opposed XYZ, it
doesn’t follow that the glass’s containing H2O has any effect on Smith’s behavior and
mentation that the glass’s containing XYZ would not have had.
 Conceptual Atomism and the Computational Theory of Mind

And it is easily shown that, indeed, Smith and Smith* coincide in respect of their
causal properties. Let C1…Cn be Smith’s causal properties. Given that Smith has those
properties, it follows that he will act just like Smith* if he is embedded in a situation
exactly like the one in which, in the scenario described a moment ago, Smith* is em-
bedded. The same thing mutatis mutandis is true of Smith*. It would therefore be ab-
surd to conclude from the differences between Smith’s behavior and Smith*’s behavior
(viz. that Smith reaches for a glass of H2O whereas Smith* reaches for a glass of XYZ)
that they have different causal powers.
This argument is basically identical with one given by Fodor (1987b). Let S be the
situation described by McDowell. Now suppose that Smith and Smith* switch places,
but that otherwise w* and our world remain the same. Let S* be the situation that results.
Obviously Smith and Smith* are going to act differently given that they are confronted
with numerically different objects. But that doesn’t mean that one of them is causally
different from the other. Two individuals are causally different just in case, were they to
switch places but all other factors were held constant, they would act differently from how
they in fact act. Smith and Smith* don’t satisfy this condition. Nor, for exactly similar
reasons, do any two individuals who differ only in respect of how their current condi-
tions originated. It follows that mental content is without causal powers to the extent
that it is to be understood along content-externalist lines. And from this it follows that
content-externalism strips mental content of causal potency and thus of existence itself.
An object’s causal properties are to be understood not strictly in terms of its behav-
ior in actual circumstances, but in terms of those aspects of its behavior that are invari-
ant with respect to counterfactual changes in those circumstances (Fodor 1987b). Mc-
Dowell’s argument involves a failure to appreciate this fact, and that is why it crumbles.
McDowell’s analysis also involves a failure to distinguish causal properties from
instances of such properties. On Monday, Jones beats Aaronson in a boxing match. On
Tuesday, Jones beats Brown in a boxing match. Jones uses the very same boxing tech-
niques on both days. So Jones’ boxing capabilities are instantiated on Monday, and
those same capabilities are instantiated again on Tuesday. Of course, what Jones is do-
ing on Monday is different from what he is doing on Tuesday: on Monday, he is defeat-
ing Aaronson; on Tuesday he is defeating Brown. But this is because some one set of
causal powers is instantiated in two different contexts.
Smith’s causal properties are instantiated in one context, and Smith*’s causal prop-
erties are instantiated in a qualitatively different context. But it obviously doesn’t follow
that the properties being instantiated are different. Each of Smith and Smith* is behav-
ing in exactly the way in which, were the two to switch places (but everything else was
held constant), the other would behave. This is consistent with their being identical in
respect of their causal properties. So if, as the content-externalist alleges, Smith and
Smith* differ in respect of the contents of their mental states, those differences have no
causal significance. Once causal properties are distinguished from their instances,
there ceases to be any temptation to say that Smith and Smith* are causally different.
 Chapter 1.  Basic concepts 

Content-externalism is the view two individuals may have different mental con-
tents even though, leaving aside facts about the origins of their conditions, those indi-
viduals are qualitatively identical. But we’ve seen that those alleged differences in men-
tal content are causally inert. In other words, there are no differences in causal properties
without differences in intrinsic properties. (How a given set of causal properties is ex-
pressed is a function of what is in the environment of the individual whose properties
we are considering. But that is irrelevant, as we have seen.) Mental properties are not
Cambridge properties; instance of them are not causally inert or explanatorily trivial. If
they are anything at all, mental properties are causal properties. So a difference in men-
tal properties must mean a difference in causal properties.18 If Smith’s mental content
really is different from Smith*’s, then Smith’s thoughts must be causally different from
Smith*’s. The supposed differences between Smith and Smith* turn out to be causally
inert, and this means that those differences don’t exist.
Let us sum up. Differences in mental content are causally inert, and therefore non-
existent, so far as they are to be understood along externalist lines. A corollary is that
content-externalism is inconsistent not only with the causal potency of the mental, but
with its very existence.
Stephen Stich (1978, 1983) sees that content-externalism is inconsistent with the
presumption that instances of representational properties can do any causal or ex-
planatory work. (As Stich puts it, content-externalism is inconsistent with the “auton-
omy of the mental.”) At the same time, Stich takes it for granted that content-external-
ism provides a correct analysis of the concept of representational content. On this
basis, Stich holds that we must reject the idea that our mental states are to be under-
stood in terms of the concept of representation or in terms of any concepts that involve
that concept (e.g. the concepts of belief, hope, doubt, desire, and so on). Stich con-
cludes that “folk-psychology” in its entirety must be thrown out and replaced with an
entirely different kind of psychology.
Stich is right to say that content-externalism strips representational properties of
any causal or explanatory significance. But Stich is guilty of a major non-sequitur in
inferring from this that folk-psychology must be jettisoned. For, as we will see, folk-
psychology is easily reconciled with content-internalism and thus with the fact that
psychological properties are causally and explanatorily important.

18. See Kim (1993: 339-367).

 chapter 2

The predicative nature of sense-perception1

Let us begin by considering a statement made by Jon Barwise (1989: 26):

When I look around I cannot see a single thing-it-itself, some sort of ideal physi-
cal object stripped of its properties and its relations with other objects. What I do
see is a scene, a complex of objects having properties and bearing relations to one
another. The properties and relations are every bit as important to what I see as the
idealized thing-in-itself. In fact, what really counts is the whole complex of ob-
jects-having-properties-and-bearing-relations which constitutes the scene.

One sees states of affairs. One does not see objects simpliciter. This must be understood
aright. Obviously one does see objects. That is a datum. But one sees objects by seeing
the states of affairs in which they are embedded. You don’t just see Fido. You see an
object with certain properties (e.g. it has beige fur and a certain number of legs) in a
certain location. That location is, in its turn, visually represented to you in relational
terms. Your seeing a given place consists in your being given visual information that
describes that place. Your seeing the place occupied by Fido consists in your being
visually told that there is a place L such that Toonces the cat is next to L and such that
L is underneath the piano, and so on.
Everything that we just said about Fido is true mutatis mutandis of each of the
objects in terms of which Fido is visually represented to you. One doesn’t just see
Toonces. One sees an object with certain morphological, chromatic, and kinematic
properties. One sees a four-legged creature with black with white spots, moving about
in such and such a manner.
Before drawing what I believe to be some of the consequences of these points, I
would like to make a distinction. Barwise rightly says that you don’t ever see a bare
thing-in-itself: whenever something is given to you in visual perception, it is given to
you as having various properties and standing various relations vis-à-vis other objects.
But there are two ways to interpret this point. We can take it to mean:
(*) You see a bare thing-in-itself plus various instances of various properties and
relations.

1. Stroud (2001) describes sense-perception as “predicative.” I am using this term in the same
way as Stroud.
 Conceptual Atomism and the Computational Theory of Mind

Or we can take it to mean:

(**) You don’t see a bare thing-in-itself at all. Your seeing a given thing consists in
instances of various properties and relations being given to you. It does not
consist in your seeing some bare thing which is, so to speak, surrounded by
various instances of properties and relations.
(*) and (**) are inconsistent with each other. If there is any possibility of a thing’s being
visually represented to one as a bare particular – even as a bare particular that happens
to be surrounded, so to speak, by various instances of properties and relations – then
it is ipso facto false to say that things must be represented as being propertied and inter-
related. So if (*) is right, then (**) is false. Given that sense-perception is necessarily
predicational, it follows that there is no prospect of a thing’s being visually represented
as a bare particular and that (**) is therefore the right thesis. In perception you are not
given a bare particular alongside instances of properties and relations. You are just
given those instances.
Sense-perceived particulars are represented to you in terms of instances of proper-
ties and relations. When you see Fido, the content of your visual perception (so far as
it can be put into words) is not given by a noun (“Fido”), but by a sentence (“at this
point in time there is, in a certain place, a creature that has four legs and beige fur…”).
For you to see Fido is for Fido to be described to you. Perception is description.
Here an anticipation is in order. Later we will find that the content of a perception
is not a proposition, even though the content of any perception can, at least up to a
point, be given by a proposition. But this doesn’t affect the substance of our points
about perception. The information encoded in any sense-perception is descriptive and
existential in nature. The content of that perception is given by information that af-
firms the existence of situations satisfying certain conditions and that, in so doing,
describes objects. A consequence is that, so far as that information can be verbalized,
it is given by existence-claims (“there is a four-legged animal with beige fur, sitting
underneath the piano…”), and not by mere nouns (“Fido”). We will often refer speak
of this information as consisting in “propositions.” But this is simply an expository
device – the word “information” can always be substituted for “proposition”, provided
the needed grammatical changes are made – and it mustn’t be supposed that we are
taking it for granted that perceptual content is propositional in nature. On the con-
trary, we will later go to great lengths (Chapters 22–23) to show that it is not, this being
one of the reasons why SCT crumbles.

Some consequences of the predicative nature of sense-perception

We’ve seen that perception is description – that a perception of an object gives one a
description of that object. This fact, taken in conjunction with a platitude of semantics,
has far reaching consequences as to the nature of representational content. That se-
 Chapter 2.  The predicative nature of sense-perception 

mantic platitude is this: expression-types must be distinguished from expression-to-

kens. The expression-type “I” doesn’t refer to anyone; but any given token of it does
refer (to the person who produced it). The expression-type “I am tired” isn’t true or
false, but particular tokens of it are.
Given these points, consider the following semantic rule:
(S) Tokens of “Smith” refer directly to Smith. In other words, there is some object
x such that x is identical with Smith and such that, in virtue of having the form
FSmith has phiG, a sentence-token is true exactly if x has phi, it being irrelevant
what other properties x might have.
At the moment, we may set aside the question whether natural languages in fact contain se-
mantic rules like (S). (Our discussion of (S) will itself help us settle that empirical question.)
The import of (S) is this. Tokens of “Smith” are not “connotative” or “descriptive.” For
a token of “Smith is tall” to be true, it is necessary and sufficient that Smith be tall; and it
is irrelevant whether Smith has any properties other than that of being tall. Put more
formally, there is some individual x such that x is identical with Smith and such that a
token t of “Smith is tall” is true iff x is tall, it being entirely irrelevant what other proper-
ties x may or may not have. So, for some x such that x is identical with Smith, a token of
“Smith is tall” has for its literal meaning the singular proposition x is tall.
But we must distinguish (S) itself from one’s knowledge of (S). First of all, (S) is a
primitive semantic rule, i.e. it is a rule that assigns meaning to an expression that
doesn’t decompose into other expressions. It must therefore be assumed that people
know (S). There can be unknown derived semantic rules. There are semantic rules of
English that assign meaning to sentences that are 27,000 words long. Nobody knows
those rules. But matters are different with primitive semantic rules. If nobody knew
that “red” meant red, then “red” wouldn’t mean red. An unknown primitive semantic
rule is no semantic rule. (Astonishingly, some philosophers, e.g. Soames 2001, have
denied that any semantic rules, whether primitive or derived, need be known. In the
course of this book, we will see many reasons to reject this view, and we will identify
the fallacies in the arguments of those who have thought otherwise.) No unknown
semantic rule can assign meaning to any sentence that people understand or to any
part of any sentence that people understand. People therefore know (S), given that they
understand sentences of the form form F…Smith…G Considering how fundamental a
knowledge of semantic rules is to communication, it is not unreasonable to expect that
a high-resolution understanding of this knowledge will shed light on the relationship
between literal and communicated meaning – between semantics and pragmatics.
Such high-resolution knowledge is not hard to obtain. We need only answer a very
basic question: How does one learn (S)? There are a few of different ways, and we will
soon discuss all of them. But right now we may focus on the most fundamental one.
 Conceptual Atomism and the Computational Theory of Mind

Smith is right in front of you. Your companion points to him and says “that guy –
the one picking up a tennisball – is Smith.” Here there is some x such that the semantic
rule that you are learning is simply this:
“Smith” refers to x [or, more precisely, tokens of “Smith” refer to x].

Now that you know (S), you can understand and meaningfully produce sentences of
the form F…Smith…G Your knowledge of (S) is thus implicated in your being able to
understand and intelligently use expressions of that form. (By an “intelligent” use of an
expression I mean one that embodies an understanding of what it means. People do,
whereas parrots do not, intelligently say “it’s cold out today.”)
But here we come to a crucial point. In order for you to know (S), you have to be
able to affix that label to the right person, and you thus have to know who Smith is, at
least on one delineation of that expression. Of course, you do know who Smith is, and
you do know whom the word “Smith” labels. You know this on the basis of the osten-
sive definition discussed in the last paragraph: you saw Smith, and were told that
“Smith” labels him. But what was the content of the visual experience through which
Smith was represented to you?
For reasons that we discussed a little while ago, you didn’t just see Smith. You didn’t
see a “bare thing in itself.” Of course, you did see Smith. But what was visually repre-
sented to you, in your seeing him, was not a bare thing-in-itself, but was rather an object
having a certain morphology (e.g. it had a certain number of limbs), certain kinematic
properties (e.g. it was picking up a tennis-ball), certain spatiotemporal properties (e.g. it
was next to a certain water-fountain near a certain tennis-court in a certain park…).
So even though (S) itself is descriptively threadbare, the information through
which (S) was made known to you was replete with descriptive content. Consequently,
you must access (S) through such information. Semantically, “Smith” simply labels
Smith. But you grasp (S) by having a mental state whose content is given by some
proposition of the form:
(#) there is an individual x such that x had such and such morphological, spatio-
temporal, kinematic…properties, and such that I saw x on such and such an
occasion, and such that “Smith” labels x.
As (#) indicates, you know that, for some x, “Smith” is a mere label of x. But given the
truth of (#), it follows that there is no x such that what a token of “Smith is tall” conveys
to you is confined to the bare proposition x is tall. What follows, in fact, is that what
such a token conveys to you has the form:
(##) there is an individual x such that x had such and such morphological, spatio-
temporal, kinematic…properties, and such that I saw x on such and such oc-
casion, and such that “Smith” labels x; moreover, x is tall.
Let us put all these points together. For some individual x, what is literally meant by a
token of “Smith is tall” is the bare proposition x is tall. But because of how you up-
 Chapter 2.  The predicative nature of sense-perception 

loaded the relevant semantic rules, what such a token conveys to you is at least ap-
proximately like (##).

Why this analysis is consistent with our modal intuitions

For the sake of argument, suppose that, as I am proposing, (##) is what is communicated
to you by tokens of “Smith is tall.” That is consistent with the fact that you know that
Smith – the person who in fact is named “Smith” – might not have had the properties in
terms of which he was initially given to you and in terms of which you therefore think
about him. He might not have been picking up a tennis ball at the relevant time. He
might never have been in that park. He might not have been named “Smith.”
For some object x, both (#) and (##) are consistent with the fact that “Smith” is a
mere label of x. Notice that, in both cases, the occurrence of the word “labels” is well to
the right of all the descriptive information regarding the circumstances relating to
your first seeing Smith. Those propositions are to the effect that there is some x who in
fact has these various properties, and that “Smith” labels x. Both of those propositions
are consistent with the idea that x – the one who in fact is picking up a certain tennis
ball in a certain park – might never have picked up a tennis ball or been to a park in his
life. Further, your knowing this last fact is consistent with the fact that you access (S)
through (#) and also with the fact that you access the meaning of a token of “Smith is
tall” through (##).

Mental content not object-involving

Let us extend these points by telling a story. In terms of appearance, Mary and Twin-
Mary are completely indistinguishable. (Here we are supposing that Mary and Twin-
Mary are actual twins, and thus live in the same world.) Thus, to know that you were
in the presence of the one as opposed to the other, you would either have to converse
with her or study her behavior for an extended period of time. You are acquainted with
both of them and thus have concepts of both. One day you see someone whom you
know to be either Mary or Twin-Mary. There is, of course, nothing in your visual per-
ception that tells you whether you are seeing the one as opposed to the other. But, let
us suppose, the person you are seeing is in fact Mary.
Here the content-externalist takes a bold position. He says that Mary herself is
part of the content of your visual perception (Burge 1982, McGinn 1986, Kaplan 1989).
Given this, suppose that, holding everything else constant, it had been Twin-Mary
whom you were seeing. In that case, says the content-externalist, your perception
would have had a different content.
 Conceptual Atomism and the Computational Theory of Mind

Let us begin by articulating the reasoning behind the content-externalist’s posi-

tion. Suppose that you see Mary (as opposed to Twin-Mary) sipping tea from a porce-
lain cup. Let V be this perception. V is veridical exactly if Mary has the relevant prop-
erties (i.e. exactly if Mary is sipping tea from a porcelain teacup …), it being irrelevant
what Twin-Mary or any other Mary look-alike is doing. Since pieces of information are
individuated by their truth-conditions, it follows that any visual perception with dif-
ferent truth-conditions wouldn’t encode the same information as V, even if the two
sense-perceptions were otherwise identical. It also follows that there is no perception
with the same content as V in any possible world where Mary doesn’t exist. Since men-
tal states are individuated by their contents, it follows that V exists only in worlds
where Mary exists. Thus, the relationship between V and Mary is constitutive. Mary is
a constituent of V’s content and her existence is therefore metaphysically necessary for
V’s existence. V is no more capable of existing in a Mary-free world than water can
exist in an oxygen-free world. This suggests that Mary herself – the flesh and blood
object, not some Fregean concept thereof – is a component not only of the content of
your visual experience, but of that experience itself. Given obvious analogues of this
argument, it follows that anything of which one has a sense-perception is a constituent
of the content of that perception and, indeed, of that perception itself. If we accept any
form of content-externalism, McGinn (1986: 15) turns out to be entirely right: the
mind is spread out all over the cosmos, the reason being that everything that you see
– every comet, every star – is a veritable constituent of your mind.
Here is what I will now argue. Mary is not a constituent of V’s content. Mary is
described by that content, but she is not a component of it. V does exist in worlds where
Mary does not. V occupies, at most, a sub-region of the region of space-time occupied
by your cranium.
Etymologically, “content” means “contained in.” The word “content” is still used in
this sense: “the contents of your carry-on bag must be securely in place before take-
off.” But, of course, visual experiences and beliefs do not contain anything, at least not
in the sense in which your wallet contains your cash. As it is used by philosophers, the
word “content” is a technical term.
Let us start with a platitude that will help fix the meaning of that term. Whatever the
word “content” means in this context, the content of an experience cannot be distin-
guished from what it tells you. (Of course, in this context, the word “tell” isn’t meant to
have any linguistic overtones.) If your visual experience doesn’t tell you that p, then p isn’t
part of the content of that experience. If your visual experience doesn’t tell you that Mary
has phi, for some property phi, then Mary is no part of the content of that experience.
A corollary is that if what your visual experience tells you is no less consistent with
Twin-Mary’s having phi than it is with Mary’s having phi – if Mary can be replaced
with any phenomenal impostor without altering what your visual experience tells you
– then we cannot say that Mary is any more a part of the content of that experience
than Twin-Mary or any other Mary-imposter.
 Chapter 2.  The predicative nature of sense-perception 

Given only these points, we can begin to see cracks in the content-externalist’s
position. Once again, consider V (your visual perception of Mary sipping tea). By itself,
does V tell you whether you are seeing Mary as opposed to Twin-Mary (or some other
Mary-impostor)? No. V leaves that quite open, as a bit of fiction easily shows. While
having V, you are told “you are seeing somebody who isn’t Mary but looks just like her.”
You say in response: “given only what my current visual experience is telling me, I
know that you are quite wrong.” Your response is a text-book case of fallacious reason-
ing. Given only what your eyes are telling you, you have no way of knowing the truth-
value of what you’ve just been told. Of course, you are in fact seeing Mary: she is the
one off of whom the relevant light-rays – the ones causing your visual perceptions – are
bouncing. But by itself V is quite silent as to whether you are seeing Mary or Twin-
Mary or some other Mary look-alike.
For you to know whether you are seeing Mary, as opposed a Mary-imposter, you
need to consult information additional to that given you by V itself. You need to con-
sider, for example, the fact that Twin-Mary hates tea and would never drink it; or the
fact that Twin-Mary is now giving a lecture in Amsterdam (whereas you are now in
Delaware); or the fact that Twin-Mary, unlike Mary, has a medical condition which
renders her incapable of holding a teacup. So by itself your visual experience doesn’t
tell you that it is specifically Mary that is in front of you drinking tea. What tells you
this is your visual experience plus background knowledge of the kind just described.
Of course, given V, you will probably not consciously have to go through an elabo-
rate string of inferences to know that you are seeing Mary. Maybe that background
knowledge operates by way of Chomskyan subpersonal inferences (this is my own view).
Maybe it does not (this is what McDowell (1998: Chapter 15), Searle (1994) and Evans
(1985: Chapter 11) would say). What is certain is that, by itself, the information encoded
in your perception is no less consistent with Twin-Mary’s being in front of you drinking
tea than it is with Mary’s having those same properties. It wouldn’t be reasonable to deny
that the relevant differential information must be found outside that perception.
The contrary position would telescope into your visual perception various pieces
of information that obviously don’t belong there, e.g. your knowledge that Twin-Mary
is now in Amsterdam. It is clearly more reasonable to say that the relevant differential
information interacts with the information that is encoded in your perception than it
is to say that it is itself a part of that information.
Before moving on, we should address one possible misgiving that might arise with
regard to the argument just given. Sellars (1963) made a compelling case that knowl-
edge of background facts is needed assign a precise representational content to phe-
nomenology. Given only the purely phenomenological aspects of your sense-percep-
tion, you could be seeing a square or a diamond, a patch of green or a patch of red, a
small nearby object or a large distant one. It is your background knowledge that as-
signs a determinate content to this phenomenology: you know that you are looking at
the object from above, and not from an oblique angle – that is why you see it as a dia-
mond, and not as a square. (The same point mutatis mutandis explains why the object
 Conceptual Atomism and the Computational Theory of Mind

appears red, not green; small and nearby, not large and far off; and so on.)2 So it might
seem that, contrary to what I just said, background knowledge is telescoped into per-

2. As I said previously, I believe that Berkeley’s (1935) argument for this is in some ways supe-
rior to Sellars: it is more lucid and more general. See Kuczynski (2007c) for a discussion of this
I would also like to point out that, given Sellars’ insight, it immediately follows that a now
extremely popular doctrine known as «intensionalism» is quite false. Intensionalism is the view
that two experiences cannot differ in phenomenology without differing in representational
content. This entails that phenomenal content is inherently and non-derivatively representatio-
nal. Thus, for any phenomenal state S, there is some representational content R such that S bears
R regardless of what the relevant subject’s background knowledge is. So given, for example, the
sensation mediating your perception of the stop-sign in front of you, the phenomenology of that
sensation completely fixes what it tells you about the world (leaving aside the strictly post-per-
ceptual inferences that you can make on the basis of that perception).
But this doctrine is plainly false. It is at least in part (though, to be sure, not entirely) be-
cause of your background knowledge that your current visual sensation tells you that there is a
medium sized object (a post with a heptagonal sign attached to it) ten feet away from you as
opposed to a mile-high object ten thousand feet away from you. In general, intensionalism is
inconsistent with Sellars’ insight that the subject’s background knowledge obviously does deter-
mine what content, if any, is borne by a given bit of phenomenology. And that doctrine is no-
thing more than a recrudescence of the epistemological foundationalism that modern epistemo-
logy, rightly convinced by Sellars’ arguments, has rejected. (Curiously, many contemporary
epistemologists accept both intensionalism and Sellarsian anti-foundationalism.)
The falsity of Fodor’s conceptual atomism follows from that of intensionalism and also,
therefore, from that of foundationalism. There is no denying that phenomenology does have
representational content. It is at least partly in virtue of the fact that I am having a visual sensa-
tion with a certain phenomenology that I see the apple as red. But, as we know from Sellars, that
phenomenology bears that content only because it operates in concert with various other repre-
sentation-bearing mental entities. What we just said about that bit of phenomenology is true
mutatis mutandis of any content-bearing bit of phenomenology. Thus, whole networks of repre-
sentations are needed to have, and thus to upload information from, any one perception. Our
perceptions do not merely generate our post-perceptual knowledge: the former, and the bodies
of knowledge implicated therein, are constitutively, and not merely causally, related to the bodies
of knowledge that we develop in consequence of our perceptions. Therefore, post-perceptual
knowledge and conception partake of the non-atomistic character of perceptual knowledge.
Thus, since intensionalism is false, so is conceptual atomism. And since intensionalism is false,
so is content-externalism, the reason being that, as we saw earlier, content-externalism is a form
of conceptual atomism.
These points show that the the Fodor-Dretske photograph-model of conception is false.
(The term «photograph-model» is due to Evans 1982.) For x to see y, it is obviously necessary
that light-rays reflected by y disturb x’s sensory surfaces in certain ways. But, given Sellars’ point,
this is not sufficient: for x to have a content-bearing, or therefore perceptual, response to those
disturbances, x’s knowledge of the conditions of perceptions, and therefore any concepts impli-
cated in that knowledge, must mediate between sensory input and representational output. This
shows that, contrary to what Dretske (1982) and Fodor (1990, 1998) hold, the relation between
the perceiver and the thing perceived is not comparable to the relation between a photograph
 Chapter 2.  The predicative nature of sense-perception 

ceptual content – that, indeed, this is typically the case. If that is right, then the argu-
ment put forth in the previous paragraph implodes.
There is an obvious fallacy here. The kind of information of which Sellars is speak-
ing is confined to facts about the conditions of perception. The fact that Twin-Mary
takes a yearly trip to Amsterdam doesn’t fall into this category. Of course, Sellars’ in-
sight does show that the information encoded in a perception isn’t neatly separable
from background information. But blurry though the line between the two may be, it
is clear that facts about Twin-Mary’s travel-habits are well on the other side of it.3
Taken by itself, what does your visual perception tell you? What is given to you in
that perception is that a person (or, at least, a person-shaped object) is sitting next to a
cat (or at least a cat-shaped object) that is on a sofa that is a certain distance in front of
you….So far as the content of your perception can be “propositionalized”, that content
is existential: there is an object x of such and such a shape sitting next to an object y of
thus and such a shape…Perceptual content is existential.
Here we must reaffirm a caveat that was stated earlier. To say that perceptual content
is existential is not to say that perceptual content consists of propositions or any other
para-sentential (or sentential) entities. Intuition resists the idea that the content of a
perception can be exhausted by any series of propositions; and later we will find demon-
strative support for our intuitions here. But given only that visual content isn’t necessar-
ily propositional, it doesn’t follow that it isn’t existential. Plainly it is existential. For even
if that content is not itself propositional, it obviously entails some proposition along the
lines of: there is a man over there, next to a car that is near a tall building…

Content-externalism and the content-internal character of entailment

Consider a world W* satisfying the following conditions. W* is just like our world ex-
cept that, at the relevant time, it is Twin-Mary, and not Mary, who is sitting on the
couch in front of you sipping tea. Let V* be the visual experience that you have under
those circumstances. Given only this much, it would seem arbitrary to say that V (as
previously defined) and V* are numerically different visual experiences; for given only
this much, it is perfectly possible that V* arises in the same way as V from the very same
physiological processes, and that (leaving aside what the content-externalist says about
those perceptions) both are qualitatively identical in all relevant respects. But the con-
tent-externalist is forced to say that V* is numerically different from V. After all, he says
that V and V* have different contents, since the one does, while the other does not, have
Mary as part of its content. So given the reasonable assumption that content-bearing

and the thing photographed. Thus, photograph-model of perception is false, and so, therefore, is
the photograph-model of conception.
3. See Fodor (1983) for a classic discussion of the “modular” nature of sense-perception.
 Conceptual Atomism and the Computational Theory of Mind

entities are individuated by their contents, the content-externalist is compelled to adopt

the dubious position that V and V* are numerically distinct.
It should be pointed out that no less an externalist than Tyler Burge (1982) insists
that mental entities have their representational contents essentially and that it is there-
fore absurd to suppose that a perception could have had a content different from the
one it actually has. On this matter, I am in complete agreement with Burge, and I will
henceforth operate on the assumption that Burge is indeed right.
The content-externalist’s position is inconsistent with basic facts about what it is
for one piece of information to entail another. The statement X and Y are the same in-
formation is equivalent with the statement for any piece of information Z, Z can be in-
ferred from X in manner M iff Z can be inferred from Y in manner M. V and V* satisfy
this condition: nothing can rationally be inferred from the one perception unless (ce-
teris paribus) it can rationally be inferred from the other; and, in addition, nothing can
be rationally inferred from the one perception in a given manner, i.e. by way of a cer-
tain series of deductions, unless (ceteris paribus) it can rationally be inferred from the
other in the very same manner.
Permit me to elucidate this last point. The proposition Frank is tall is different
from the proposition Mary is short because there are propositions that can be inferred
from the one that cannot be inferred from the other. In general, for P and P* to be dif-
ferent propositions, it is sufficient that there be some proposition P** that can be ind-
ferred from the one but not the other. But it is not necessary. The proposition 1+1=2 is
distinct from the the proposition triangles have three sides, even though neither propo-
sition entails anything not also entailed by the other. But those propositions still differ
in respect of their entailment-relations. Given any proposition P that is implied by
both triangles have three sides and 1+1=2, the way that the one implies P is different
from the way that the other does so. Supposing that one’s starting point is that 1+1=2,
one need only apply the rule of existential generalization a single time to prove that, for
some n, n+1=2. But that is not the case if one’s starting point is that triangles have three
sides. So even though 1+1=2 and triangles have three sides coincide in respect of what
they entail, they differ in respect of how they entail it, and they therefore differ in re-
spect of their entailment-relations, that being why they are numerically distinct propo-
sitions. In general, propositions differ just in case their entailment-relations differ.
Given obvious extensions of the argument just presented, the same is true of non-
propositional pieces of information, supposing that there are such things. (This topic
will be revisited in Chapter 11.)
As we’ve seen, anything that can reasonably be inferred from V can also reasona-
bly be inferred from V*, and vice versa. And whatever can be inferred from the infor-
mation encoded in the one perception is obviously to be inferred in the same way from
the information encoded in the other perception. It follows that V and V* are identical
in respect of what they entail, and also in respect of how they entail it. Therefore, it
cannot coherently be said that V and V* encode different information. Thus, the con-
tent-externalist is wrong to say that V and V* have different contents.
 Chapter 2.  The predicative nature of sense-perception 

The content-externalist’s typical response to this sort of point is as follows:

If the content-externalist is right, V tells you that Mary is now sipping tea, where-
as V* tells you that Twin-Mary is now sipping tea. So this last argument of yours
simply begs the question against the content-externalist.

But we’ve already seen why this point is a non-starter, as it involves a failure to distin-
guish the information encoded in our perceptions from the information that, with the
assistance of background knowledge, we derive from our perceptions.
A related point is that the content-externalist’s position involves a failure to distin-
guish between content and truth-maker. V is a perception of Mary, as opposed to
Twin-Mary, not because Mary is a constituent of V’s content, but because some fact
about Mary is V’s truth-maker. As we’ve discussed, V tells you not that Mary is sitting
in a certain place sipping tea, but that some woman (or, at any rate, some entity with
the morphology characteristic of a woman) has those properties. There is indeed a
woman satisfying that description and, under the present circumstances, that woman
is Mary. But there are counterfactual scenarios where some fact about Twin-Mary
would be the truth-maker of that same content.

The relationship between causality and content

For a visual experience to be a perception of x, x must be part of the state of affairs that
causes that experience. But given only that x satisfies that condition, it doesn’t follow
that x per se is a constituent of the content of that experience. The content of a visual
experience is, as we’ve emphasized, indistinguishable from what that experience tells
you. And what that experience tells you is indistinguishable from what you can ration-
ally infer from it, taken by itself. By itself, no visual experience can tell you that it is x
specifically – as opposed to some x-look-alike – that caused it; there is no way that, by
itself, any perception could encode such differential information. This is easily estab-
lished on generically logical grounds.
What your visual experience, taken by itself, tells you is a function solely of facts
about the physical properties of the situation right in front of you. (Indeed, it is going
to be a function of a small subset of those facts.) So if one of constituents of that situa-
tion is replaced with something numerically distinct but qualitatively identical, there
is no way for your experience to register that substitution. Suppose that x1, ….xn are all
qualitatively identical but numerically distinct objects and that, at this moment, you
are seeing exactly one of those objects. The character of your visual experience reflects
only those aspects of the situation that are preserved in the light-rays that hit your
retinas; and, for any i and any j, those aspects are invariant with respect to intersubsti-
 Conceptual Atomism and the Computational Theory of Mind

tutions of xi and xj. Thus, your visual experience, taken by itself, cannot possibly regis-
ter the fact that it is x3, as opposed to x24, that you are seeing.4
There is thus no way that your visual experience, by itself, could discriminate be-
tween its being Mary in front of you and its being Twin-Mary in front of you. So if we
plausibly identify the content of your visual experience with what that experience tells
you; and if we plausibly identify what it tells you with what you may logically infer
from it (or, at least, with a certain proper part of what you may logically infer from it);
then we must say that Mary is no more a part of your visual experience than Twin-
Mary or any other Mary-impostor. Given this last point, if we are to avoid theoretical
arbitrariness, we must say that none of Mary or Twin-Mary or Robo-Mary…is a con-
stituent of that content. All of this is consistent with the fact that your perception is of
Mary specifically, and all of this is consistent with the fact that Mary – or, rather, some
state of affairs involving Mary – is necessarily the cause of any perception of Mary.

Content-externalism and disjunctivism

We can show how radical and counterintuitive the content-externalist’s position is by

following his own reasoning to its logical conclusion.5 Once again, let V be a visual
perception of Mary sitting on a couch holding a teacup. The essence of the content-
externalist’s view is that Mary herself is a constituent of V’s content. But, of course, the
same is supposed to be true each of the individual objects in the situation that you are
seeing. So if content-externalism is right, the just mentioned sofa is a veritable compo-
nent of V’s content, and so is the cat sitting on that sofa – and so on.
Given this, let V1 be a perception that is just like V (as defined a few pages back) with
this one qualification: the sofa represented by V has been replaced with a qualitatively
identical, but numerically distinct, sofa. Now let V2 be a perception that is just like V1

4. Russell (1927: 94) makes a similar point when he writes: “Prejudice tells you that you see
the same [Russell’s emphasis] table on two different occasions; you think that experience tells
you this. If it really were experience, you could not be mistaken; yet a similar table may be subs-
tituted without altering the experience…there is nothing to show that one identical entity causes
the two sensations.” See also Russell (1948: 271-294).
5. See Evans (1982: chapter 5). In his discussion of Evans (1982), Rick Grush writes: “Evans
then discusses the appropriate way to specify the content of informational states. His primary
point is that in specifying the content of an informational state, one need not, and one should
not, mention any specific objects. Rather, it should be specified by mentioning properties and
objective-variables (in an open sentence). This does mean that we cannot perfectly well speak of
the information being of this or that specific object. Two identical photographs might be such
that one is of Bob and the other is of Rob, Bob’s twin brother. But in specifying the content of the
photograph, we should not mention Rob or Bob, but should rather say something like “a tall
person with a red hat”, a specification that both photographs would share.” See the chapter of
Grush (1999) entitled “Guide to Chapter 5 of Gareth Evans’ Varieties of Reference.”
 Chapter 2.  The predicative nature of sense-perception 

except that the teacup represented in V1 has been replaced with a qualitatively identical,
but numerically distinct, teacup. Repeat this procedure until each of the objects repre-
sented by V – each bit of yarn, each carpet-fiber – has been replaced with a qualitatively
identical, but numerically distinct object, and let Vn be the resulting perception.
The similarity between V and Vn is not purely phenomenological. An artist who
experienced V and painted what he saw would ceteris paribus produce a painting qual-
itatively identical with that produced by an artist who experienced Vn and then painted
what he saw. Further, there is some one description D such that, in virtue of having
either V or Vn, one would be warranted in affirming D. (D would be along the lines of:
there is a woman with such and such properties sitting on a sofa holding a teacup…) It is
clear that V and Vn have a great deal of representational content in common.
But content-externalism cannot accommodate this obvious fact. For the content-
externalist, every one of the replacements we described would constitute a change in
representational content. Numerically different objects, qualitatively different contents.
A corollary is that two perceptions have the same content only to the extent that the
objects represented by the one are numerically identical with those represented by the
other. So, according to the content-externalist, V and Vn have no content in common.
But intuition recoils at such a view and, as we’ve just seen, it is demonstrably false.
John McDowell recognizes that, if content-externalism is right, then V and Vn
have no representational content in common. Because he is a content-externalist, he
accepts that V and Vn have no common content. This position, appropriately general-
ized, is sometimes referred to as “disjunctivism”, since it is given by the thesis that the
content of the one perception is completely “disjoint” from the content of the other
(McDowell 1998: Chapters 10–13).
Of course, McDowell concedes that V and Vn are identical in terms of phenome-
nology. But, in his view, that is where the similarity ends. McDowell does not address
the fact that V and Vn have a great deal of existential information in common and that,
consequently, the content of the one does at least overlap with that of the other.
A content-externalist might take the position that V and Vn do have some content
in common – that they have a purely “qualitative” or “general” content in common.6 So
each of them would be to the effect that some object having such and such properties is
sitting on a sofa…But, at the same time, V would be to the effect that X is sitting on sofa
Y, whereas Vn would be to the effect that X* (≠X) is sitting on sofa Y*(≠Y). The idea
would be that representations have both a purely “qualitative” or “identity independ-
ent” content, and that they also an “identity-dependent” content.7
This view is either false or amounts to nothing more than a proposal that we make
what would inevitably be a fruitless and confusing change in terminology. What the

6. See Blackburn (1984: 317-322). I should make it clear that Blackburn rejects hard-line
content-externalism. Indeed, many of our criticisms of content-externalism are anticipated by
Blackburn (1984: Chapter 9).
7. This terminology is borrowed from Blackburn (1984).
 Conceptual Atomism and the Computational Theory of Mind

content-externalist is referring to as “identity-dependent content” is identical with

what we have seen to be, not content, but meta-content. A related point is that the
“content” that V and Vn don’t have in common is of a very different kind from the con-
tent that they do have in common. The latter has inferential and causal powers that the
former lacks. (In virtue of having different identity-independent contents, two visual
perceptions or beliefs have different causal powers, and they also warrant different
inferences, But, as we have seen, two perceptions or thoughts that have different iden-
tity-dependent contents do not necessarily differ in respect of what they cause or what
can be inferred from them.) Consequently, the externalist is using the word “content”
to denote two very different things and, once this equivocation is exposed, his view
loses whatever plausibility it might initially appear to have.
I would like to address a possible objection to the analysis just outlined (I will put
it in the voice of an imaginary spokesperson for content-externalism):
As you yourself have insisted, if a perception, or any other content-bearing entity,
has content C, then it necessarily has that content. If B is someone’s belief that
snow is white, then it is nonsense to suppose that B might have been a belief that
1+1=2. Given this point, it obviously follows that, if p is a perception of Mary, it is
necessarily a perception of Mary. After all, in this context, it is a matter of simple
semantics that the verbiage to the right of the word “of ” denotes the content of the
perception in question. In English, if I wish to indicate that X is a constituent of
the content of some perception of mine, I say that I had a perception “of ” X. Deny-
ing this would be like saying that “Socrates” doesn’t denote Socrates. For reasons
that you yourself have given, it follows that Mary herself, as opposed to some
Fregean sense thereof, is a constituent of the content of that perception.

On a number of occasions, philosophers have drawn extravagant epistemological and

metaphysical conclusions on the basis of innocent facts about language.8 This is one of
those cases.
In almost all contexts, what is of interest to others, and even to yourself, is not what
is encoded strictly in your visual experience, but is rather what you know on the basis of
that experience. Most of what we wish to say would be virtually incommunicable if our
locutions were faithful to the division of labor characteristic of our information-gather-
ing. Suppose that language forced one to make it clear exactly what was due to perception
and what was due to extra-perceptual background knowledge. In other words, suppose
that if one wished to explain how one knew that Mary had gone to the airport, one had
to produce a statement of the form:
(#) “I had a visual experience such that what was given to me in that experience
was an individual x having a distinctively female morphology… and such

8. The stock example of this is Meinong’s view that “the square circle” and “Pegasus” do refer
to objects, albeit ones that “subsist” and do not “exist.” See Russell (1905).
 Chapter 2.  The predicative nature of sense-perception 

that, given my background knowledge, it was evident to me that Mary unique-

ly satisfied the just stated existence-claim.”
In that case, one would need extreme cognitive and linguistic virtuosity to make and,
to a lesser extent, to understand even the most pedestrian of statements. In narrow
philosophical contexts, (#) is an informative and relevant statement. But in most con-
texts, the stratifications that (#) so faithfully preserves are of no interest. What we want
to know is whether you have information about Mary (or Fred or Ethel…). The micro-
structure of the cognitive processes by which you attained that knowledge is irrelevant
and, given the philosophical and psychological obstacles involved, would also be ex-
traordinarily difficult to learn about in the first place.
For this reason, language does us a great service by telescoping (#) into “I saw Mary”,
or “I had a perception of Mary”, and by effecting other, comparable condensations across
the board. Were it not to do this, language could be used and understood only by a few
prodigies, and could convey only in the most circuitous manner the information that all
of us (including the aforementioned prodigies) typically wish to know. So given only
that I must report what I saw by saying “I had a perception of Mary” (or, more idiomat-
ically, “I saw Mary”), it doesn’t follow that Mary is a constituent of the content of the
visual experience itself. The primary purpose of language is to convey what is pragmati-
cally (not philosophically) relevant. There is thus no reason to expect that our locutions
will always reproduce the delicate distinctions to which we must look for solutions to
recherché questions of logic and epistemology. Indeed, given the practical nature of lan-
guage, it is to be expected that language will obscure those distinctions.
 chapter 3

Uniquely individuating descriptions

We’ve seen a number of reasons to accept the following three views:

(1) Conception is essentially descriptive in character.
(2) Expressions like “Socrates” and “Plato” are not descriptive to any degree: they
are “directly referential.”
(3) There is no difficulty reconciling (1) and (2) with each other.
In his landmark work Naming and Necessity, Kripke made a strong case for (2). (Like
most contemporary philosophers, I believe that Kripke’s arguments are decisive.1) On
this basis many have rejected (1), believing it to be incompatible with (2). In this chapter,
I would like to provide reasons additional to those already put forth to believe that (1)
and (2) are compatible with each other.
Let us start with a quick summary of some of the relevant points in Naming and
Necessity. According to Russell (1918, 1948), “Socrates” is synonymous with a definite
description; this might be “the greatest philosopher to have died of hemlock poisoning.”
To understand the word “Socrates”, and thus to understand sentences of the form F…
Socrates…G, one must obviously know the semantics of “Socrates”; and one must there-
fore know that “Socrates” is synonymous with that definite description. Consequently,
anyone who uses the word “Socrates” with understanding is able to identify a descrip-
tion that is satisfied by Socrates and Socrates alone. In other words, such a person knows
some true and non-trivial proposition of the form: Socrates uniquely has phi. (An exam-
ple of a trivial proposition of that form would be: Socrates uniquely has the property of
being identical with Socrates. An example of non-trivial proposition of that form would
be: Socrates was uniquely a greatest philosopher to die of hemlock-poisoning.)
In general, if one knows the semantics of a referring term, then one knows some
description that applies uniquely to the referent of that term. (By “applies to”, I mean
“is satisfied by.” I will also use “singles out” as a synonym for both of these locutions.)
Supposing that x is the referent of “Gilgamesh”, if I understand what is meant by sen-
tences like “Gilgamesh was wise” or “Gilgamesh lived a long time ago”, then I know of
some description that singles out x. If I know of no such description, then I cannot
really know what is meant by such sentences.
Supposedly, Kripke demolished this view. Consider any description that Socrates
uniquely satisfies. Here is an example: “the greatest philosopher to have died of hemlock-
poisoning.” One can understand sentences of the form F…Socrates…G without knowing

1. See Soames (2001, 2005).

 Conceptual Atomism and the Computational Theory of Mind

that the referent of “Socrates” consumed hemlock. One can therefore understand “Soc-
rates was wise” without knowing that “Socrates” picks out somebody who uniquely satis-
fies the description: greatest philosopher to have died of hemlock-poisoning.
This point (mutatis mutandis) holds for any description that singles out Socrates.
Consider the description: “the person who figures as the protagonist in most of the
platonic dialogues.” One can understand the sentence “Socrates was wise” without
having the slightest idea that he was a protagonist in any philosophical dialogues. Giv-
en any description D that applies uniquely to Socrates, one can understand F…Socra-
tes…G without knowing that D applies to the referent of “Socrates.”2
This point has many apparent consequences. Some of these are semantic in nature.
Others are epistemological.
Here is one of the semantic consequences. “Socrates” is not synonymous with any
definite description. Given any description D (with a few trivial exceptions), one can
know that “Socrates was wise” is true without knowing that somebody uniquely in-
stantiated D and that any such person was wise. (The trivial exceptions have to do with
descriptions like person who was identical with Socrates. More explicitly, they relate to
descriptions the grasping of which presupposes a concept of Socrates.) So it isn’t tenable
to maintain that “Socrates” is synonymous with “the greatest philosopher to have died
of hemlock-poisoning” or with any other definite description.
Here is one of the supposed epistemological consequences. One can think about
Socrates without knowing of any description that singles him out – without knowing a
uniquely individuating description (UID) of the kind just described.

The causal theory of reference and conception

Kripke’s argument has other apparent consequences, both for semantics and episte-
mology. Let us start with a question: Why does “Socrates” refer to Socrates and not (for
example) to Plato? Russell’s answer was this: “Socrates” is synonymous with a descrip-
tion that is true of Socrates and Socrates alone. Kripke made a powerful case against
this answer. So how does Kripke answer our question? His answer is that “Socrates”
refers to Socrates because a certain kind of causal relation holds between tokens of
“Socrates” and Socrates himself (or, more exactly, states of affairs involving Socrates).3
The following story will help us further explicate Kripke’s views on this topic.
Smith personally saw Socrates. (We may set aside as irrelevant the fact that none of
Socrates’ contemporaries were named “Smith.”) Socrates’ presence thus caused Smith
to have certain visual experiences. Smith then gave a name to Socrates. (Let us idealize
away from the fact that, for cultural reasons, that is probably not how Socrates was
named.) Smith was able to pass that name along to people who did not themselves see,

2. Kripke (1972: Lecture I).

3. Kripke (1972: Lecture II).
 Chapter 3.  Uniquely individuating descriptions 

or otherwise sense-perceive, Socrates. Of course, for Smith to pass along that name to
someone, that person had to be causally connected to Smith and, therefore, to Socra-
tes. The process just described is repeated over and over. Let us suppose that I am the
521st installment in a sequence of events like the one described.4
Smith’s original term for Socrates may well have undergone phonetic distortion.
What I have been left with is the sound (or sound-type) “Socrates”, which may be very
different in terms of phonetics from the term that Smith used. But that is not relevant
here. What is relevant is that, when I produce instances of that sound-type, my words
refer back to Socrates precisely because I am connected to Socrates by way of a chain
like the one just described. When I say “Socrates”, my utterance refers to Socrates be-
cause it is causally connected to him in a certain way. In general, if an expression E
refers to an object O, that is in virtue of the fact that tokens of E stand in a certain
causal relation to states of affairs involving O.

Why the causal story is only a chapter in the descriptive story

Let us now evaluate the various claims just put forth. How do you learn what a given
word refers to? At first, there appear to be two ways: you can learn it by description or
by ostension. (Soon we will see that there is a third.)
Let us illustrate these points with a story. You are young, and have thus never
heard the name “Socrates.” Somebody says to you:
(*) “Socrates” is the name of the most famous philosopher to die of hemlock poi-
soning.
Here the word “Socrates” has been defined for you in a descriptive manner. In effect,
you were told that somebody O uniquely satisfies a certain description (viz. most fa-
mous philosopher to die of hemlock poisoning) and that “Socrates” names O.
On the face of it, there seems to be a second, entirely distinct way of learning what a
term refers to: you can be given an ostensive definition. Here is an illustration. You are in
a crowded market place. You and a companion are watching a man haranguing a crowd.
While pointing to that man, your companion says:
(**) That person over there is named “Socrates.”
But, in a certain class of cases, one is able to acquire the ability to use a referring term
without ever having been given either an ostensive or descriptive definition of it. Sup-
pose that you and two other people (Smith and Jones) are talking. All of a sudden Smith
and Jones start talking about somebody named “Argo.” They never say who “Argo” re-
fers to. They never say: “Argo was the president of our fraternity back in college” or
“Argo was my divorce lawyer; and by sheer coincidence he’s Smith’s racquetball part-

4. Kripke 1972: Lecture II.

 Conceptual Atomism and the Computational Theory of Mind

ner.” And they do not ostensively define “Argo”, since he is not present. Smith and Jones
simply prattle on about “Argo”, and you are too polite to ask them who that person is.
Nonetheless, as Gareth Evans (1985: Chapter 1) pointed out, it seems that, even
under these circumstances, you can use the word “Argo” in a meaningful way. You can
say “it sounds as though Argo is a text-book case of somebody with borderline-per-
sonality disorder”; and you can ask questions like “did Argo mean to do that to you, or
was it just an accident?”
So, on the face of it, it seems that you can acquire the ability to use “Argo” in a
meaningful way without ever having been given an ostensive or descriptive definition
of that term. When you say “it sounds as though Argo is an evil person”, you are using
the expression “Argo” with understanding: you are not in the same category as a parrot
that mindlessly produces those same sounds. It thus appears that you can simply “pick
up” the ability to use “Argo”, and names generally: there needn’t be an ostensive or de-
scriptive definition.5
Let us take stock. There seem to be three ways that a proper name N can be in-
ducted into one’s lexicon:
(i) one can be given an ostensive definition of N;
(ii) one can be given a descriptive definition of N; or
(iii) one can just “pick up” the use of N.
It is easily shown that that one’s learning what an expression refers to always involves
one’s learning some description that singles out its referent. This is because all of (i)-
(iii) collapse into (ii).

5. Of course, one could say this:

Because you haven’t been given any kind of definition of “Argo” you cannot really use that term with un-
derstanding, and you cannot really know what is meant by sentences containing it. When you produce the
sounds “Argo sounds like a bad person”, you are really making some kind of metalinguistic statement, some
statement like: “ ‘Argo” refers to somebody who is bad.” And when you hear sentences like “Argo was a cad”, you
cannot really understand them; what you understand is some related meta-linguistic claim like: “ ‘Argo” refers
to somebody who was a cad.”
This position cannot be summarily dismissed; and I will not dismiss it. But, for our purpo-
ses, it is enough to note that the objector’s point seems quite false. Cases like the one just descri-
bed are pervasive. There are many cases where no ostensive or descriptive definition is involved
in our appropriation of a proper name. We hear the term being used and – presto! – can use it
ourselves. If the objector’s point is correct, we are not really using any of these terms: we are
merely mentioning them. When we produce sentences containing them, we are talking not about
people, but about symbols denoting people. But this seems false. I have never been given a des-
criptive or ostensive definition of the term “Charlemagne.” But it certainly appears as though I
can make statements about Charlemagne, and not just about “Charlemagne.” I am thus going to
operate on the assumption that I can make statements about Charlemagne himself.
 Chapter 3.  Uniquely individuating descriptions 

Let us start by considering category (i). Suppose that somebody ostensively defines
“Socrates” for me. I see somebody haranguing a mob. Pointing to Socrates, my com-
panion says: That person is named “Socrates.”
Under this circumstance, I know that there is some individual x such that x
uniquely has certain properties, and I know that “Socrates” refers to x. Ostensive defi-
nitions work by way of knowledge of uniquely individuating descriptions. (Let us now
talk about Fred, a hypothetical contemporary of ours, instead of long-gone Socrates.)
Somebody points to some person (or object) O and says: That guy is named “Fred.” In
order for the definition to work – i.e. in order for you to learn to what (or whom)
“Fred” refers – you must have some way of singling out the person ostended. Suppose
your companion points in a certain direction and says to you “that is Fred”, but there
are thirty different equally salient objects. In that case, the definition won’t work, since
you aren’t able to pick out the right object.
Now suppose that there are thirty numerically different, but (leaving aside their
locations) qualitatively identical men in a certain place. Your companion Bob goes up
to one of them, puts his hand on his shoulder, and says:
(*) This guy is named “Fred.”
Even though the person just named is qualitatively just like many others, you know of
some description that applies to him and to no one else. For you know the following
proposition to be true: There is some individual x such that x is uniquely a person whom
Bob just picked out (i.e. such that x uniquely has the property that Bob just picked x out);
moreover, “Fred” names x. This time Bob’s definition works. It works only because,
thanks to it, you know of some existence-claim that Fred, and Fred alone, satisfies.
The description is not one that tells you much about what kind of person Fred is. In
fact, that description doesn’t tell you anything about how Fred is psychologically or
(leaving aside the information is encoded in your perception) physically different from
other people. The description in question gives you enough information to cognitively
single out Fred, and it gives you nothing beyond that. But that is neither here nor there.
In this context, there is only one important point, namely: your knowing what “Fred”
refers to involves your knowing the truth of some proposition of the form: somebody x
uniquely has phi and “Fred” names x.
This is not to deny that your knowing what “Fred” refers to involves your having
some kind of causal connection to Fred. You do have some kind of causal connection
to Fred: you see him, and this involves your being causally connected to him. You ac-
quire the ability to use the word “Fred” in consequence of this causal connection, and
your subsequent uses of “Fred” thus refer to Fred in virtue of this same connection.
But the causal connection is a component of your knowledge of a uniquely individu-
ating description. In the story just told, that description was (at least approximately):
there is some object x such that x is over there and such that a moment ago Bob put his
hand on x’s shoulder and said that “Fred” is x’s name. The relevant causal connection
 Conceptual Atomism and the Computational Theory of Mind

between you and Fred corresponds to the context-sensitive terms in that statement.
(These are the expressions in italics.)
We will see that in any case where a causal connection between X and Y is consti-
tutive of X’s concept of Y, that connection is embedded in X’s knowledge of some de-
scription that Y uniquely satisfies. We will also see that, if that description were verbal-
ized, the just mentioned causal connection would be expressed by one or more of the
context-sensitive expressions in that verbalization.
Let us now deal with category (iii). We will soon see why we are skipping to (iii),
and not dealing right away with (ii).
Once again, consider the scenario, described a moment ago, where you are talking
with Smith and Jones. Remember what we said earlier about that situation. Smith and
Jones are acquainted with Argo, but you are not. And at no point in the conversation
do they ever say anything like “Argo was my divorce lawyer” or “Argo was the best man
at my wedding.” They never say of some definite description that it refers to Argo; they
never say of some uniquely individuating description that Argo satisfies it.
Of course, given that they are producing sentences of the form F…Argo…G, it fol-
lows that you are being apprised of many indefinite descriptions that he satisfies. When
Smith says “Argo is a real cad”, you are being given an existence-claim of the form:
somebody x is a cad and “Argo” refers to x. But, of course, many people are cads. So
given only that you know the truth of this existence-claim, it doesn’t follow that you
know to which specific person – which specific cad – “Argo” refers.
But the truth is that, in virtue of the fact that you overheard Smith and Jones using
the word “Argo”, you have been given a definite description that Argo uniquely satis-
fies, and you have learned some true proposition of the form: something x uniquely has
phi and “Argo” refers to x. Indeed, you have been given many such descriptions and
many such propositions.
We may show this through an extension of our story. Unlike you, Smith is person-
ally acquainted with Argo. For reasons that we’ve already covered, Smith thus knows
some true proposition of the form: something x uniquely has phi and “Argo” refers to x.
Let us now take the next step in our argument. On each occasion where Smith or
Jones says “Argo was a cad”, there was one person to whom Smith was referring by
means of the word “Argo.” Suppose that, on occasion t, Smith says “Argo was a cad.”
Obviously the description Fx is a cadG is indefinite, and applies to many different indi-
viduals. But on that occasion, there was exactly one person whom Smith was describ-
ing as being a cad; and you know this merely by virtue of overhearing Smith’s state-
ment. So just by virtue of being in colloquy with Smith, you do pick up knowledge of
some claim having the form on occasion t, there was exactly one person whom Smith
described as being a cad, and “Argo” refers to x.
We must be careful to distinguish between the following claims:
(a) Exactly one person has phi.
 Chapter 3.  Uniquely individuating descriptions 

(b) Exactly one person is being described, on some particular occasion, as having
phi.
Your picking up the word “Argo” doesn’t involve your learning the truth of some claim
of the form: exactly one person x is a cad, and “Argo” refers to x. Such a claim is false,
since there are many cads. Your picking up “Argo” involves your learning the truth of
some claim of the form: on such and such occasion, exactly one person x is being de-
scribed as a cad, and “Argo” refers to x.
We must take care to make one other distinction. Smith himself never uttered an
existence-claim of the form: there is some x such that x uniquely has phi and “Argo” re-
fers to x. Smith himself never uttered an existence-claim of the form: on occasion t, I
described somebody x as a cad and “Argo” refers to x. Smith said only “Argo is a cad.” But
the information that you are given, by virtue of being in discussion with Smith, is not
confined to what Smith himself says. When Smith says “Argo is a cad”, the information
that is encoded in his words is quite threadbare. But even though what Smith was say-
ing might have been minimal, the information that you have about Smith’s saying of it
is enormous. You know when and where Smith performed that speech-act, and you
have much other non-trivial information about it: and that information does suffice to
give you the right kind of uniquely individuating description.
There is no vicious circularity in this account.6 Smith personally met Argo. He
learned the meaning of “Argo” through an ostensive definition. Maybe Smith himself
gave Argo the name “Argo.” (For cultural reasons, this is not likely. But that is not rel-
evant.) In any case, Smith saw Argo, and then learned (or possibly stipulated) that the
person he was seeing was named “Argo.”
Let us sum up. You meet Smith. You have a conversation with him. On some spe-
cific occasion within that conversation, Smith says “Argo was a cad.” You thus know
that, on that occasion, there is somebody x such that Smith is describing x and x alone
as a cad, and that “Argo” is the term Smith is using to refer to x. The uniquely individu-
ating description that you subsequently associate with “Argo” is not: the unique person
x such that x is a cad named “Argo.” Rather, it is: the unique person x such that, on such
and such occasion, Smith was describing x as a cad.
In the circumstance just described, you are only at one remove from Argo, since you
are speaking with somebody to whom “Argo” was defined ostensively. But what we’ve
said applies equally to the case where you are at five hundred removes from the referent
of the name in question. Consider the following scenario. Smith doesn’t personally know
Argo. But Smith knows Brown, and Brown personally knows Argo. Smith is not given an
ostensive or descriptive definition of “Argo.” But Smith overhears Brown saying “Argo is
a cad.” On this basis, Smith is able to add “Argo” to his own lexicon. (We just described
the principles governing such an addition.) You then overhear Smith saying “Argo is a
cad”, “Argo is a mean person”, and so forth. You are not given an ostensive or descriptive

6. It is not guilty of the circularity that Kripke (1972) rightly ascribes to one of Russell’s theo-
ries concerning the semantics of proper names.
 Conceptual Atomism and the Computational Theory of Mind

definition of “Argo.” But on the basis of what you overhear, you are able to add “Argo” to
your lexicon: the principles that govern Smith’s appropriation of “Argo” obviously apply
to your appropriation of that name. The same principles would apply if you were five
hundred removes from somebody who was actually acquainted with Argo. Invariably,
what would enable you to use “Argo” to refer to Argo would be your knowledge of some
existence-claim that Argo uniquely satisfies.7
Let us now consider category (ii). Here it is very easy to fall prey to a confusion.
Imagine the following. You have never heard the word “Socrates” before. You are talk-
ing with your friend Charlie. He says “Seymour is second only to Socrates in philo-

7. As we discussed earlier, words undergo phonetic degradation as they change hands. The
first person who uses the word “Argo” might pronounce it one way; the people who “pick up”
that name from that person might pronounce it differently. But this fact obviously doesn’t affect
the principles governing the appropriation of that word. Brown might pronounce “Argo” one
way; Smith might pronounce it differently; you might pronounce it a third way; the people to
whom you transmit this word might pronounce it in yet other ways. After a few transmissions,
what is left might bear little resemblance to Brown’s pronunciation. In some contexts, that fact is
of importance. But that fact has no bearing on our point that, whenever somebody “picks up” a
name – whenever they add it to their lexicon in manner (ii) – it is in virtue of their acquiring
knowledge of a uniquely individuating description of the kind that we have been describing.
It is obvious how these points apply to names like “Socrates.” The word that Socrates’ own
acquaintances used to refer to him was not “Socrates” (although that word probably had some
phonetic similarities to “Socrates”). Nonetheless, however distorted “Socrates” may be relative to
the word that the Ancient Greeks used, the principles governing the “picking up” of that word
are the same. In each case, whenever one adds “Socrates” to their lexicon in manner (ii), it is in
virtue of their acquiring knowledge of the right kind of uniquely individuating description.
Suppose that “Sukrat” is the Ancient Greek word for Socrates. Glaucon learns what “Sukrat”
means through an ostensive definition. Glaucon is engaged in a discussion with Zenon. Glaucon
never gives Zenon any ostensive or descriptive definition of “Sukrat.” But, in Zenon’s presence,
Glaucon does utter a sentence of the form F…Sukrat…G Thus, for reasons discussed, Zenon acqui-
res the ability to use “Sukrat” to refer to Socrates; and, for reasons already discussed, Zenon’s
ability to do this involves his knowing a uniquely individuating description of the previously
discussed kind. Zenon has a conversation with Phoebus. Zenon slightly mispronounces “Sukrat.”
In a manner analogous to that just described, Phoebus acquires the ability to use that mispro-
nounced version of “Sukrat” to refer to Socrates. In this way, variants of “Sukrat” are transmitted
from person to person, and from generation to generation. What we are left with, two thousand
years later, is a word that is quite different (at least phonetically) from “Socrates.” But one’s acqui-
ring the ability to use that word, however distorted it might be relative to the original word for
Socrates, involves one’s learning some true claim of the form: somebody x uniquely has phi and
“Socrates” names x.
In most cases, phi will be some property like thing described, on such and such occasion, by
so and so, as having thus and such property. “Phi” will not be a property of historical moment; it
will not be a property like great philosopher of antiquity who drank hemlock or philosopher who
figured as the protagonist in most of Plato’s dialogues. The property in question will be some hi-
ghly contextual one, some property that doesn’t connect in any significant way with Socrates’
intellectual or historical countenance.
 Chapter 3.  Uniquely individuating descriptions 

sophical ability.” Because you don’t know who “Socrates” refers to, you ask: “Who is
Socrates?” Charlie says: “Socrates was the author of the Nichomachean Ethics and the
Posterior Analytics.”
Of course, Charlie is wrong. Socrates did not write either of those works, and “Soc-
rates” does not refer to the author of either of those works. At the same time, in conse-
quence of your having this exchange with Charlie, you can use the word “Socrates” to
refer to Socrates.
Here it seems as though you are able to use the word “Socrates” to refer to Socrates,
even though your only belief about him is wrong – even though the only thing you
“know” about him isn’t true. But this appearance is misleading. What we have here is
a case where you add a word to your lexicon in manner (iii). You do know some
uniquely individuating description that applies to Socrates. But that description is not
“the unique author of the Nichomachean Ethics and the Posterior Analytics.” Rather,
it is: Fthe person to whom, on occasion t, Charlie ascribed the property of writing the
Nichomachean Ethics and the Posterior AnalyticsG, for some value of t.
In most cases where we seem to be appropriating a name in manner (ii), we are in
fact appropriating it in manner (iii). But we can certainly construct a case where, un-
mistakably, we are appropriating a name in manner (iii).8
Suppose you and Charlie are having a conversation about the lives of pioneers in
the area of clothing-fashions. Neither of you has any idea who invented the zipper. You
and Charlie decide that, at least when speaking to each other, you will refer to that
person, whoever it is, as “Julius.” (See Evans 1985: Chapter 7.) So you have no idea who
invented the zipper; but you know that somebody uniquely did so, and you have stipu-
lated that “Julius” is to refer to that person. Thus you have added a semantic rule to
English (or, at any rate, to a certain idiolect of English). It must be made clear what that
rule is. The rule is given by the following:
(Z) If x uniquely invented the zipper, then “Julius” refers to x.
The descriptive information is given wide-scope with respect to the “refers to” opera-
tor. So if Smith was a unique zipper-inventor, then “Julius was a millionaire” means
Smith was a millionaire. That sentence does not mean: somebody x uniquely invented
the zipper and x was a millionaire. “Julius” is a name, not a quantifier.
You and Charlie don’t know who “Julius” refers to. But it does refer to somebody.
On the face of it, it certainly seems as though you can use it significantly: you can, with
understanding, say things like “Julius was probably a talented inventor” and “Julius
must be worth a fortune.”
Some authors, e.g. Kaplan (1968) and Donnellan (1974), have said that, under the
circumstances, you cannot say things like “Julius must be worth a fortune” and really
mean them. This is controversial. (I will soon argue that it is true.) But what is clear is

8. Here I am basically repeating a point made, in different forms, by Kripke, Kaplan, and
Evans.
 Conceptual Atomism and the Computational Theory of Mind

that if you and I can use the term “Julius” with understanding, it is only because we
know some true claim of the form: somebody x uniquely has phi, and “Julius” refers to
x. (In other words, a knowledge of such a claim is a necessary, though perhaps not a
sufficient, condition for being able to understand utterances of that kind.) Here, of
course, phi is the property of inventing the zipper.
I do not wish to be misunderstood. A case can be made (and soon will be) that
unless you know who invented the zipper, you don’t know which proposition is seman-
tically encoded in “Julius must be worth a fortune”, or in any other sentence of the form
F…Julius…G What you actually know is some meta-linguistic proposition like: for some
object O, the sentence “Julius must be worth a fortune” is true exactly if O must be worth
a fortune. Since you don’t know which exact person invented the zipper, you don’t
know to whom “Julius” refers, and you therefore don’t know which proposition is en-
coded in that sentence. But, at the same time, you have metalinguistic knowledge about
“Julius must be worth a fortune” which, in some respects, simulates a real understand-
ing of it.
In everyday life there are many cases where we seem to be adding a name to our
lexicon in manner (ii). Suppose that little Timmy has never heard the name “Shake-
speare” before. You say to him: “Shakespeare is the person who wrote Romeo and Juli-
et.” Here it seems that little Timmy is appropriating the name “Shakespeare” in manner
(ii); for what you are telling him is, in effect: there is some x such that x uniquely wrote
Romeo and Juliet and “Shakespeare” refers to x.
But really little Timmy has appropriated “Shakespeare” in manner (iii). A Kripke-
style argument shows this. After your conversation with him, little Timmy can make
statements about Shakespeare. They may be wrong and ill-informed. But they are
wrong and ill-informed statements about Shakespeare. Let us suppose that Timmy
produces the words “did Shakespeare like French fries?”, and “Shakespeare must have
been weird, because his writing is so funny.” (I say “produces the sounds”, instead of
“asks” and “says”, so as not to prejudge the question whether Timmy is, in fact, saying
or asking anything.)
But suppose it turns out that, although he wrote Hamlet and King Lear and Timon
of Athens, Shakespeare did not write Romeo and Juliet. Does that mean that little Tim-
my’s Shakespeare-statements were blanks? Does it mean that, in asking “did Shake-
speare like French fries?”, Timmy wasn’t really asking anything? Does it mean he was
merely simulating what is done by somebody who is actually asking a question? It
seems not. It seems that little Timmy did manage to ask whether Shakespeare liked
French fries.9
But if Little Timmy’s ability to refer to Shakespeare depended on Shakespeare’s
having written Romeo and Juliet, then his statements would indeed have been abortive.
Little Timmy couldn’t ask or say anything about Shakespeare if his ability to refer to

9. This directly contradicts Evans’ (1982: Chapter 7) position. Chapter 8 of the present work is
dedicated to discussing whether Evans’ view is correct or not.
 Chapter 3.  Uniquely individuating descriptions 

that man depended on the truth of the statement: somebody uniquely wrote Romeo and
Juliet and “Shakespeare” refers to x.
What is going on here is that little Timmy added “Shakespeare” to his lexicon in man-
ner (iii). He heard you refer to him. Thus, given what we said earlier, the existence-claim
through which he added “Shakespeare” to his lexicon has the form: there is some x such
that, on such and such occasion, Daddy said that x wrote Romeo and Juliet and “Shake-
speare” refers to x. That existence-claim remains true, even though the other one – some-
body x uniquely wrote Romeo and Juliet and “Shakespeare” refers to x – turned out false.
Let us now talk about the existence-claim that turned out to be wrong. What role
did this proposition have? Another piece of fiction will help us answer this question.
You see somebody at a party. You don’t know him. He acts as though he is very
important. You ask your companion: “Who is that guy? What gives him the right to be
so self-important?” You are told: “He is the dean.” Before you knew who that person
was, you were able to refer to him, and also to have thoughts about him. You were able
to think that he is obnoxious and to wonder what gives him the right to be so haughty.
But, despite this fact, there is still some sense in which you don’t know who he is. You
can think about and refer to somebody without knowing who they are. When you say
to little Timmy: “Shakespeare is the guy who wrote Romeo and Juliet”, you are telling
him (or trying to tell him) who Shakespeare was: and this, as we’ve just seen, is differ-
ent from transmitting to Timmy the ability to think about or refer to Shakespeare. (Of
course, what we’ve said about “Shakespeare” is true of proper nouns generally.)
Most “descriptive definitions” are really cases where one is learning who some-
body is, or which object something is. They are not cases where one is acquiring the
ability to refer to, or think about, that thing. Usually, but not always, the ability to refer
to, and think about, the object in question seems to pre-exist knowledge of who some-
body is or which object something is.
So while there may be cases where one is genuinely appropriating a name in man-
ner (ii), they are not easy to produce, and they tend to be highly artificial. The only
possible examples arise in connection with expressions like Evans’ “Julius.”
Let us summarize what we have found thus far. Any case of one’s adding a refer-
ring term to one’s lexicon involves one’s knowing some description that singles out the
referent. For any name N, one’s adding N to one’s lexicon involves one’s knowing some
true claim of the form: somebody x uniquely has phi, and N names x. This is true even
in the case where one acquires the name through ostensive definition.
The existence-claims are not usually ones of historical moment; they are not ones
like somebody x wrote Romeo and Juliet and “Shakespeare” names x. They are usually
highly specific to the peculiarities of one’s own situation. But what is important is that
knowledge of some such claim, trivial though it may be from a historian’s standpoint,
is needed to use a name to refer to a thing.
 Conceptual Atomism and the Computational Theory of Mind

Neo-descriptivism untenable

Even though we learn what a name refers to through some existential-descriptive claim,
the name itself is just a label. (In any case, this is what Kripke’s arguments strongly sug-
gest, and we will continue to find confirmation for this position.) Semantically, “Shake-
speare” simply labels Shakespeare. It doesn’t have any descriptive content. But, as we’ve
seen, one always learns which thing a name labels through some existential-descriptive
claim. The semantics are not existential-descriptive; but the information involved in
the process of learning the semantics is existential-descriptive. The semantic rule for
“Shakespeare” must be given to you in some way or other. But what is given must be
distinguished from how it is given.
The following statement is false:
(S1) The semantic rule for “Shakespeare” is this: somebody x uniquely wrote Ham-
let and “Shakespeare” names x.
(S1) is false because “Shakespeare” refers to Shakespeare in worlds (that are identical with
ours in respect of the semantic facts that hold in them) where Hamlet was never written
and where, consequently, the semantic rules of English do not presuppose that Hamlet
was ever written.
Unlike (S1), the following statement is true:
(S2) Somebody x uniquely wrote Hamlet and, and the semantic rule for “Shake-
speare” is this: “Shakespeare” names x.
Notice that, in (S2), the expression “the semantic rule for ‘Shakespeare’ is this” is given
narrow scope with respect to the information that somebody wrote Hamlet. This is
how it must be. Otherwise we end up committed to the kind of descriptivism that
Kripke refuted. “Shakespeare” is just a label. For some x, the literal meaning of “Shake-
speare was prolific” is simply: x was prolific. Nothing about the authorship of Hamlet is
to be found in the literal meaning of that sentence.
Though false, (S1) doesn’t involve as many falsehoods as the following:
(S3) The semantic rule for “Shakespeare” is this: for any predicate F…x…G F…
Shakespeare…G means: somebody O uniquely wrote Hamlet and…O…
There is some x such that what is literally meant by “Shakespeare was bald” is simply:
x was bald. According to (S3) “Shakespeare was bald” means: somebody x wrote Hamlet
and x was bald. Therefore (S3) is wrong. Why is (S3) more wrong than (S1)? According
to (S1), there is some x such that an utterance of “Shakespeare was bald” means simply
x was bald. So (S1), unlike (S3), does say what is meant by “Shakespeare wrote Hamlet”,
even though, for the reasons given a moment ago, (S1) is false.
To avoid running afoul of Kripke’s correct points, we must take care to give the
relevant operators – Frefers to__G, Fis the semantic rule for__G – narrow scope. We must
make sure that they are safely to the right of potential descriptive, and therefore modal,
 Chapter 3.  Uniquely individuating descriptions 

contaminants. We’ve seen that whenever you learn the semantics of a name, it is
through descriptive information. But you give the descriptive content wide-scope. That
is why, if you are told that proper name “N” refers to the thing which happens to be the
unique phi, you don’t regard FN has phiG as analytic, or as true in all possible worlds.
The semantic information is given narrow-scope; the epistemological information is
given wide-scope; and that is why you have all the right modal intuitions.
Dummett (1973: 110–151) argues for wide-scope descriptivism. In his view, “Soc-
rates was necessarily a philosopher” means (or is logically equivalent with): there was
some x such that x was a unique great philosopher to drink hemlock and necessarily: x
was a philosopher. Dummett’s view, appropriately explicated, is as follows. “Socrates”
has a Fregean sense, this being (let us suppose) the concept philosopher who drank
hemlock. Thus, in virtue of containing an occurrence of “Socrates”, a sentence encodes
(inter alia) the existential proposition that something uniquely falls under that con-
cept. But given such a sentence, any modal expression occurring in it is given narrow-
scope with respect to that existential information. Thus, “it is necessarily the case that
Socrates was a philosopher” has for its meaning the proposition: somebody x is
uniquely a philosopher who drank hemlock, and it is necessarily the case that x was a
philosopher. So, for some x, what that sentence describes as necessarily true is the sin-
gular proposition: x was a philosopher. Of course, that singular proposition is not nec-
essarily true, since no one has to be a philosopher. Dummett’s analysis therefore cor-
rectly predicts that “it is necessarily the case that that Socrates was a philosopher” is
false. Given any sentence in which a proper noun falls within the scope of a modal
operator, similar remarks show that Dummett’s analysis correctly identifies the modal
status of that sentence. In this way, Kripke’s compelling modal arguments are recon-
ciled with Frege’s compelling semantics.
It is clear that this doctrine is false10, the reason being that “Socrates was bald”
doesn’t entail “somebody drank hemlock.” Dummett’s analysis deals with some of
Kripke’s modal points, but not with Kripke’s compelling point that there is no analytic
connection between “Socrates was a philosopher” and “somebody drank hemlock.”
It is to be emphasized that (S2) is true only because, in it, the expression “the se-
mantic rule for ‘Shakespeare’ is this” is given narrow scope, so that no descriptive in-
formation falls within its clutches. By being careful about where we put that operator,
we avoid any kind of semantic descriptivism, and our analysis remains Kripke-friend-
ly without absurdly requiring that conception be non-descriptive.

10. See Soames (2001) for an excellent discussion of Dummett’s descriptivism.

 Conceptual Atomism and the Computational Theory of Mind

The problem of forgotten start-up descriptions

There is a point that we must address. It will help if we first define a bit of terminology.
Let
(^) something x has phi
be a true proposition. Of course, what makes it true is that some particular thing O has
phi. The truth of any generalization supervenes on the truth of some singular proposi-
tion. Let us refer to O as the “verifier” of (^).
Consider the sentence “somebody is bald.” This sentence is true because particular
people are bald – because Bob and Fred are bald. Bob and Fred are therefore verifiers
of “somebody is bald.” An existence-claim can have many verifiers. (False existence-
claims have none.) A true existence-claim of the form “somebody uniquely has phi”
has exactly one verifier.
With this terminology in place, we can easily state a formidable objection to our anal-
ysis. One typically forgets the existence-claim through which one added a name to one’s
lexicon. One typically forgets the first time one heard “Shakespeare” or “Aristotle”; one of-
ten forgets one’s first meeting with a person. Given this, one might be tempted to say:
You don’t necessarily associate any uniquely individuating description with “Aris-
totle” or “Smith” or “Jones.” Such a description may have been necessary to enable
you to refer to those people and also to think about them. But one those abilities
are in place, the descriptions drop out. At that point, there can be pure thought
about the entities in question. You can just think about Aristotle and Smith; you
don’t have to think about them through a uniquely individuating description.11

Obviously one does often forget the existential information that “started up” one’s con-
ception of some object – one often forgets such “start-up” descriptions, as we might call
them. But we must not, on this account, accept the objector’s position. It is hard to believe
that conception is as impoverished, as barren of descriptive content, as the objector
makes it out to be. I thus propose an alternative view, namely: when one forgets a start-up
description, what happens is not that one ceases to have knowledge of a uniquely indi-
viduating description, but rather that one replaces one such description with a related
but distinct one.
Remember what we said earlier about “Fred.” If you add “Fred” to your lexicon in
manner (i), i.e. through an ostensive definition, that involves your knowing some claim
of the form: somebody x uniquely has phi and “Fred” names x. Given this, let us tell
another story. You and a friend are at a party. You see somebody talking animatedly.
That person is wearing a bowler hat. At time t, your friend points to that person and

11. This point was made in private correspondence (in response to an earlier draft of the present-
work) by Steven Horst. Horst’s position is found in Fodor (1998), Burge (1982) and, indeed, in al-
most any work that makes any concessions to content-externalism.
 Chapter 3.  Uniquely individuating descriptions 

says “that is Fred.” Your adding “Fred” to your lexicon involves your knowing some
existence-claim – some proposition (or piece of information) that is at least approxi-
mately of the form:
(e1) there is now somebody x such that I am now sense-perceiving and focusing
on x, and such that x is wearing a bowler hat; moreover, “Fred” names x.
Let us suppose that, a minute or so after “Fred” is ostensively defined for you, you go up
to him and start talking. You now have a new uniquely individuating description of him.
There are a number of reasons for this. First of all, the mere lapse in time between your
current situation – your now talking with Fred – and the initial ostensive definition by
itself guarantees that you will have a new uniquely individuating description of him. At
time t, e1 was your uniquely individuating description. But because that time has now
passed, e1 has been replaced with:
(e2) there is somebody x such that I was sense-perceiving x a minute or two ago,
and such that x was then wearing a bowler hat; moreover, “Fred” names x.12
Notice that e2 “interlocks” with e1. The thing described by e2 is identical with the thing de-
scribed by e1. Of course, the uniquely individuating description changes. But the new one
is related to the old one in such a way that something x uniquely makes both of them true,
and the new one thus inherits the verifier of the old one. In a moment we will say more
about this process of inheritance or transmission.
But factors other than the mere passage of time force e1 to be replaced with a new
uniquely individuating description, as the following continuation of our story makes
clear. You are now engaged in conversation with Fred. You can see him more clearly.
From afar he looked hale and strong, but you now see that he is puffy and out of shape.
He is saying a number of things to you (he is talking about the national debt and the
current administration). As a result, you generate a new uniquely individuating de-
scription of Fred, namely
(e3) there is somebody x such that I was sense-perceiving and focusing on x a
minute or two ago, and such that x was then wearing a bowler hat and such
that I am now conversing with x, and such that x is now talking about the
national debt, and such that x looks puffy and unhealthy; moreover, x is
named “Fred.”
Notice that e3 and e2 interlock. e2 transmits its verifier to e3.
Now that you have e3, you can drop e2 and e1. You can completely forget them, and
that will not jeopardize your ability to think about Fred. Suppose that a moment pass-
es and that you have stopped talking with Fred. By itself, the mere passage of time re-
places e3 with a slightly different claim:

12. Kaplan (1989) makes a similar point.

 Conceptual Atomism and the Computational Theory of Mind

(e4) there is some x such that, a moment ago, x was talking to me about the na-
tional debt, and such that x looked puffy and unhealthy; moreover, “Fred”
names x.
Suppose that you have completely forgotten e1-e3. You still know of some description
that Fred uniquely satisfies. This will continue to be the case so long as you are able to
think about Fred. You go home at night. e4 is now replaced with:
(e5) there is some x such that, several hours ago, at a party that I attended, x was
talking about the national debt, and such that x looked puffy and unhealthy;
moreover, “Fred” names x.
This process can continue ad infinitum. One’s ability to think about and refer to some-
thing must be started up by one’s acquiring knowledge of the right kind of existence-
claim. Typically, that specific claim is forgotten. But, if so, it is replaced by another ex-
istence-claim that interlocks with it. That second one may, in its turn, be replaced by a
third; the third by a fourth; and so on. Thus, contrary to what the objector says, the
ability to think about and refer to an external object always involves knowledge of an
existence-claim – a uniquely individuating description – that singles it out.

The concept of recognition

The objector says that there can be “pure”, i.e. non-descriptive, conceptions of objects.
But conception is surely not so innocent. This becomes clear when we consider the
phenomenon of recognition.
To recognize something is not, at least not merely, to register the truth of a bare
identity proposition – a proposition of the form O=O. To register the truth of Smith is
Smith is not, by itself, to recognize anything.
Recognition is obviously replete with knowledge of descriptive content. When you
recognize the man in front of your house as the man who insulted you at the restaurant,
a great deal of descriptive, and therefore existential, knowledge is involved. You know
that there is some x such that x insulted me at the restaurant…You also see that there is
some y such that y is now standing in front of my house…Your recognizing the man in
front of your house as the man from the restaurant involves your “integrating” these
pieces of information. You see that both descriptions – man in front of my house and
man from the restaurant – connect somehow. More specifically, you see that there is
some one thing such that both descriptions apply to it. In other words, you recognize the
truth of some claim like: there is some x such that x is a man who insulted me at the res-
taurant and there is a man y such that y is now standing in front of my house, gesticulating
threateningly; moreover, x is identical with y.
Once it is granted that recognition involves putting together different pieces of
existential-descriptive information, it becomes highly plausible to see recognition as
 Chapter 3.  Uniquely individuating descriptions 

consisting in one’s registering the truth of the kind of existence-claim just described.
The alternative is to see recognition as stripped of any descriptive content. But this is
not psychologically or epistemologically plausible.
These points about recognition confirm our analysis of conception. When you rec-
ognize something or someone, what is happening is that two conceptions converge.
Your recognizing the man now standing in front of your house as the belligerent lout
from the restaurant involves your integrating two different conceptions; it involves your
realizing that some one conception (the awareness you are having now, of the man in
front of your house) and some other conception (the one you have of some ruffian
whom you dealt with at a certain restaurant) both pick out some one thing. If concep-
tions were “pure”, i.e. non-descriptive and therefore non-existential, then the just men-
tioned convergence of conceptions would be stripped of any descriptive content. Since
descriptive information wouldn’t be involved, recognition would be reduced to the reg-
istering of some bare claim of the form O=O. But surely that is not what recognition is.
Surely your recognizing so and so at least involves your seeing that so and so has certain
features or properties. Given the role that conceptions have in recognition, it thus be-
comes necessary to ascribe some of this descriptive knowledge to conceptions. It thus
becomes necessary to see conceptions as involving knowledge of existence-claims of
the kind just discussed. This fact, combined with the fact that we often forget the de-
scription which starts up a conception, gives plausibility to the analysis we gave a mo-
ment ago regarding the sustaining of a conception: a conception is sustained by knowl-
edge of a series of interlocking existence-claims.

The causal analysis as a component of the descriptive analysis

None of this belittles the role that causal connections play in conception and refer-
ence. I am able to think about Fred precisely because I saw him. And, of course, my
seeing something involves that thing’s affecting me in some way.
But how does that causal connection help make conception possible? What enables
me to think about Fred is my knowledge of some uniquely individuating description
(there is some guy x such that I am now seeing and attending to x, and such that x is wear-
ing a bowler hat; moreover, “Fred” names x). As we discussed earlier, my seeing Fred in-
volves my being apprised of the truth of an existence-claim that he uniquely satisfies. In
other words, my seeing Fred involves my being given a description of him. My being
causally connected to Fred is involved in my being given that description. The causal con-
nection is internal to the description. That causal connection is preserved if that particu-
lar description is forgotten and replaced with one that interlocks with it, in the manner
described a moment ago. At any point, my conception of Fred is constituted by my know-
ing a description that is either identical with, or descended from, that encoded in a cer-
tain perception. Either way, what is doing the work is my knowledge of a uniquely indi-
 Conceptual Atomism and the Computational Theory of Mind

viduating description of Fred. The causal connection, though essential, is a constituent of

my knowledge of that description.

Appendix to Chapter 3
The similarities and dissimilarities between our analysis and Russell’s

According to Russell, for me to have a concept of O, I must know the truth of some
existence-claim of the form x uniquely has phi that O satisfies. So, like us Russell, advo-
cates a strictly descriptive conception of conception (except where sense data and uni-
versals are concerned).
But, on the basis of this correct view, Russell put forth some doctrines that are high-
ly questionable. The purpose of this section is to determine, in light of the points thus far
developed, what is right about Russell’s position and also what is wrong about it.
Russell famously distinguished between knowledge by description and knowledge
by acquaintance.13 According to Russell, one’s knows something by description if one
knows some existence-claim that it uniquely satisfies; and one knows something by
acquaintance if one has some kind of non­-descriptive and, therefore, completely direct
awareness of it. (I am acquanted with my own pains; but I have, at most, knowledge by
description of other people’s pains. My knowledge of my own pains cannot be analyzed
into some more fundamental kind of knowledge. But my knowledge of other people’s
pains can be so analyzed. So far as I know about another person’s pain, it is because I
know that, given that person’s behavior, there must be some internal state having cer-
tain phenomenological properties – it is because, in other words, I know some exist-
ence-claim that describes that pain. My knowledge of my own pains is, of course, not
so indirect.) Russell thus concludes that one cannot be acquainted with any constituent
of the external world. The only spatiotemporal entities with which one can be ac-
quainted, according to Russell, are bits of one’s own immediate consciousness (e.g.
sensations, images).14
For this reason, Russell comes to the conclusion that it isn’t possible for anyone
(with the possible exception of Socrates himself) to grasp propositions of which Soc-
rates is a constituent. Russell believes that if x is an external spatiotemporal entity – a
spatiotemporal entity that is not a constituent of one’s own consciousness – then one
cannot grasp any singular proposition of the form x has phi. Put another way, one can-
not grasp any proposition that de re concerns Socrates or any other external object. I
will argue that ultimately Russell is wrong. But the operative word here is “ultimately”;
and the reasons why Russell is now generally held to be wrong have little or no real
force against his views.

13. See Russell (1917: 152-167).

14. At some junctures of his career, Russell held that one can be acquainted with universals and
also with one’s self. See Russell (1984).
 Chapter 3.  Uniquely individuating descriptions 

What is right about Russell’s view

Consider the singular proposition:

(*) Socrates was bald.
To grasp (*), one must presumably grasp each of its constituents. In particular, one
must have a concept of Socrates. But for reasons that we’ve already touched on, to have
a concept of Socrates is to know some existence-claim that he uniquely satisfies. So one
grasps (*) by way of grasping some proposition of the form:
(**) Something x uniquely has (or had) phi, and x was bald.
There is no such thing as a “pure”, descriptively innocent grasp of anything external. It
follows that (*) cannot be grasped except by way of (**). This is obviously similar to
Russell’s view. Russell chose to express this view by saying that one simply cannot grasp
(*). I choose to express it by saying that the way one grasps (*) is by grasping (**). But
this difference is purely verbal.15
Russell made a point that has not been given its due in contemporary epistemol-
ogy. There does seem to be a fundamental difference between the way I know about my
own pains and tickles and the way I know about external objects. Suppose that I am
having some pain x, and at that very moment, I believe true the singular proposition x
is unpleasant and I wish it would go away. It doesn’t seem possible that, if x were not to
exist, I would be capable of having that very thought. After all, x is a constituent not
only of the proposition that I am grasping, but also of my grasping of it.
By contrast, an external object can never be a constituent of one’s awareness of it.
Sense-perceived objects are no exception to this principle. A story may help make this
clear. Tom is in possible world W1 and he sees Socrates. V1 is the purely psychological
component of Tom’s perception. Socrates doesn’t exist in world W2. But W2 is other-
wise just like W1. So supposing that V2 is Tom’s visual experience W2, it follows that V1
and V2 are numerically identical. Of course, the distal cause of V1 is Socrates, whereas
the distal cause of V2 is (let us say) some android who resembles Socrates. But the im-
mediate biological antecedents of V1 are numerically identical with immediate bio-
logical antecedents of V2. More generally, both experiences are embedded in the same
way in the history of the same organism. It would thus be highly counterintuitive to
deny that they were numerically identical. 

15. Russell also held, at least at certain junctures, that every proposition that one grasps is equi-
valent with some proposition about sense-data. This, everyone (including myself) now believes,
is quite wrong. But the position that one’s beliefs about the external world are beliefs about
sense-data is entirely distinct from the position that one grasps external objects by knowing the
truth of existence-claims that they uniquely satisfy. It is one thing to hold that singular proposi-
tions can be grasped only by way of general ones. It is another thing to hold to hold that singular
propositions concerning external objects can only be grasped by way of propositions concer-
ning one’s own sense-data.
 Conceptual Atomism and the Computational Theory of Mind

Also, V2 and V1 are identical in respect of their phenomenologies and also in re-
spect of their causal properties. (Some content-externalists would say that they have
different causal properties. But we’ve already seen why that position is a non-starter.)
Let us now use the word “perception” to denote the kind of phenomenologically
pregnant experience that, if veridical, would have a real-world object. So, in this con-
text, a drunkard has “perceptions” of pink elephants no less than you are having per-
ceptions of this book.
Given what we said a moment ago, it is easy to demonstrate the following two
principles. First, given any perception P, if O is the object of that perception, there is a
possible world where P exists even though O does not. Second, given any instance A of
some propositional attitude, if O is one of the objects of A (i.e. if O is one of the things
that A is about), then (so long as O is not itself one of the mental or neurological enti-
ties composing A), there is a possible world where A exists even though O does not.
(In this context, the variable “O” is to range only over spatiotemporal entities – it is not
to range over platonic abstracta.)
There is a third principle that falls out of our discussion. (Although it is, I believe,
a tautological consequence of the previous two principles, it is nonetheless worth stat-
ing separately.) Let M be some awareness – it could be a perception or a thought – that,
in actuality, concerns O, where O is not one of the neural or cognitive events that real-
ize M. Being a spatiotemporal entity, M has various causal properties (certain things
cause it to occur and it, in its turn, causes certain other things to occur). and M is obvi-
ously individuated by at least a certain subset of those causal properties. Given this,
here is the just mentioned third principle: supposing that C1…Cn are the causal prop-
erties that, in actuality, are individuative of M, there is a possible world where O does
not exist but where there is some entity M* that has all of C1…Cn.
Permit me to say what this last principle amounts to. Let M be any mental entity
that, in actuality, represents object O. Supposing that O is not itself one of the events
composing M, there is a possible world where M exists and O does not, and there is a
possible world where O does not. In other words, where there exists something having
all the causal properties that, in actuality, are individuative of M.
Put simply, unless the object O of an awareness A is actually a component of that
same awareness – unless O is one of the mental or physiological occurrences that make
up that awareness – A could exist in the absence of O. And the reason is that, unless O
is such an object, O’s existence is not constitutively involved in A’s having the causal
properties that individuate it.
So if I am thinking about or seeing Socrates, there is no essential, and thus no
counterfactually resilient, relation between my awareness and Socrates. That same
awareness could exist in Socrates’ absence.
The externalist is likely to respond to this by saying the following:
If, upon seeing Socrates, you say “that guy looks tired”, your words are either true
or false, the same being true of the perception that motivated them. But if, upon
 Chapter 3.  Uniquely individuating descriptions 

having a hallucination qualitatively just like that perception, you say “that guy
looks tired”, your words are neither true nor false, the same being true of the pseu-
do-perception that motivated them. Thus, the contents of hallucinations have
blanks in them that, in the contents of phenomenologically identical veridical
perceptions, are filled by actual objects. So, contrary to what you say, there are
awarenesses that one simply couldn’t have in a Socrates-free world. In such a
world, one could have, at most, a perception that was phenomenologically identi-
cal with a Socrates-perception but whose content had either a gap in it or, instead
of having such a gap, comprised somebody other than Socrates.

It is not hard to identify the fallacy in this argument. Contrary to what the objector says,
a hallucination does tell you something true or false. For it tells you something false. It
tells you that, in a certain place, there is an individual having certain properties. Since
your hallucination would not (at least not as a whole) tell you anything true or false if
there were a blank in its content, the objector is wrong to say that there is such a blank.
Indeed, a veridical perception of Socrates would (ceteris paribus) have for its content
the very same information as the just described hallucination. It follows that, in a Soc-
rates-free world, one could have a perception (or hallucination) that had the same con-
tent as a veridical perception (had by somebody in this world) of Socrates.
It is true, that when, in an attempt to refer to some hallucinated man, the halluci-
nator says “that guy looks tired”, his words don’t encode anything true or false, where-
as those same words may encode a truth (or a falsehood) when uttered by somebody
who has a phenomenologically identical veridical perception. For, as Kaplan (1989)
showed, given an utterance of “that guy looks tired”, the relevant semantic rule is:
If there is a unique contextually salient guy x, then a token t of “that guy looks
tired” is true iff x looks tired. If there is no such an individual, then t is abortive,
i.e. it doesn’t encode any proposition.

But that is irrelevant in the present context, given the massive gulf between literal
meaning and cognitive content.
Externalists are also likely to wage the following attack on our argument:
Perceptions, and awarenesses generally, are individuated by their objects. What
could be more obvious? A perception of Socrates couldn’t be anything other than
a perception of Socrates – anymore than Socrates himself could have been Plato
or Napoleon.

We’ve already seen, at length, why this line of thought is untenable.

Now we can see the power of Russell’s position. Let A be some awareness of mine,
and let O be the object of that awareness. As we’ve been emphasizing, if O is not a
component of A, then there is a counterfactual world where A exists but O does not.
But if O is a component of A, then (for at least some values of O) there might well be
no counterfactual world where A exists but O does not. And, in that case, O’s existence
 Conceptual Atomism and the Computational Theory of Mind

is necessary for A to have the causal properties that make it what it is. An illustration
might be appropriate.
Suppose that you are having a headache and you think to yourself: I wish this head-
ache would go away. Let T be that thought (in this context the word “thought” refers to
a certain kind of mental entity), and let H be the headache in question. It seems rea-
sonable to say that T couldn’t possibly exist in a world where H did not exist. In con-
nection with this, it seems that H is not only the object of T, but that it is itself a com-
ponent of T. So given any world w there H doesn’t exist, there isn’t anything identical
with T. This is because, given any world where H doesn’t exist, there is nothing that is
enough like T in respect of its genetic and, more generally, causal properties to be T. If
the line of thought just presented is correct, then Russell was right to say that bits of
one’s own consciousness can be veritable constituents of the propositions that one
grasps. As we saw earlier, he was also right to deny that the same could be true of ob-
jects in the external world. Russell’s epistemology thus seems to be vindicated.
Let us summarize the argument just presented. Spatiotemporal events are individu-
ated by their causal properties – but not by all of those properties. Any two non-simulta-
neous events ipso facto have some causal connections. Given any event E involving some
brontosaurus, E has some causal relation to everything I do. But there are epistemically
possible worlds where E does not occur but where my life is, in every significant respect,
exactly as it is here. So spatiotemporal states of affairs are individuated by a certain priv-
ileged subset of their causal properties. (I will not attempt to give a precise definition of
that subset.) In light of these points, let A be an awareness that has some external object
O for its object, and let C1…Cn be the causal properties individuative of A. There is a
possible world where O doesn’t exist but where something x has all of C1…Cn, i.e. there
is a world where A exists but where O does not. So A’s existence is only extrinsically con-
nected with that of O. Further, for reasons that we have discussed at length, in a world
where O doesn’t exist but A does exist, A will have exactly the same truth-conditions –
the same content – that it has here. The context-externalist’s view to the contrary, as we
saw, involves a failure to distinguish between, on the one hand, the data encoded in A
and, on the other hand, meta-data that concerns that data.
But now consider an awareness A* that represents an object O* such that O* is itself
a constituent of A*. There is (for at least some values of O*) no world where A* exists
but O* does not. An awareness has this sort of counterfactually resilient relationship to
its object when, and only when, that object is a constituent of that awareness. Finally, in
any world where O* does not exist, there is nothing that has quite the same truth-condi-
tions as A*. Thus, it is reasonable – possibly even de rigueur – to hold that the data en-
coded in A* itself has O* itself as a constituent and, consequently, that A*’s relationship
to O* is fundamentally different from the relationship that we have seen my concept of
Socrates to bear with respect to Socrates.
If the last three sentences are correct, then Russell’s position is quite correct. After
all, on at least one reasonable understanding of the term “content”, the content of an
awareness is internally related to that awareness. We’ve just seen that any perception of,
 Chapter 3.  Uniquely individuating descriptions 

or thought about, Socrates could exist, and have the very content that it has in actuality,
in worlds where Socrates does not exist. (Of course, this point holds for any awareness
A having an object O that is not itself a constituent of A).
Given this, Russell is right to say that the data encoded in one’s thoughts or percep-
tions never has Socrates per se as a constituent. After all, any thought that did have Soc-
rates himself as a constituent obviously couldn’t exist in counterfactual scenarios where
Socrates didn’t exist. So, to rephrase the point just made, Russell is right to say that any
thought about, or perception of, Socrates is not de re with respect to Socrates (except
perhaps in cases where the thinker is Socrates himself – but probably not even then, for
reasons having to do with personal identity that we cannot discuss here).

What is wrong with Russell’s view

Russell holds that one cannot be acquainted with singular propositions among whose
components is a constituent of the external world. This is not true. Because my argu-
ment for this claim is a complicated one, let me summarize it before providing it in full.
Let P be an arbitrary proposition that has as one of its components a constituent of the
external world. (So P might be the singular proposition: Socrates was wise.) There is
some proposition P* that meets the following two conditions. First, P* is purely de-
scriptive: its only constituents are properties and relations. Second, P* completely de-
scribes the external states of affairs on which the truth of P supervenes. Because P*
meets the first condition, one can grasp it (at least in principle). Because P* meets the
second condition, there is nothing that P tells one about the external world that P*
does not tell one. (In fact, P* provides much more information than P.) It follows, I
believe, that there must be an internal relation between P* and P – a relation of iden-
tity or inclusion. In a word, singular propositions are purely descriptive propositions
in disguise; and since one can (in principle) grasp the latter, one can also grasp the
former. As we will see, one is not often acquainted with such a proposition. In fact,
acquaintance with such a proposition is probably a practical impossibility, and might
even be a nomic impossibility, given interference-effects and the like. But, contrary to
what Russell argues, it is not logically absurd to suppose that one might be acquainted
with such a proposition. Let us now state the argument just outlined.
We have some inclination to think of objects – e.g. rocks, trees, people – as being the
basic constituents of the spatiotemporal world. This viewpoint is encouraged, though
probably not wholly caused, by the fact that expressions denoting objects are semanti-
cally simple. Among the ultimate, semantically simple expressions composing a sentence,
one typically finds expressions like “Socrates”, “Plato”, and the like. And in cases where
expressions denoting objects have semantic complexity – e.g. “that rock over there”, “the
inventor of bifocals” – that semantic complexity serves the purpose of picking out the
relevant individual, not (at least not typically) of delineating its structure. So expressions
denoting objects are either semantically simple or what they complexity they have has
 Conceptual Atomism and the Computational Theory of Mind

nothing to do with the structure of the relevant object. And this surely reinforces a pre-
existing tendency to think in terms of, and cognitively divide the world into, objects.
But, of course, rocks, trees, and people are not the ultimate constituents of reality. To
be sure, the term “ultimate constituent of reality” is a vague one, and whether x falls in its
extension depends on what the relevant criterion of simplicity is. (Given any object, it
may either be complex or simple, depending on how it is described and depending also
on the explanatory context. This shows that complexity is ultimately a property of propo-
sitions or explanations, and not of objects.) But it can safely be said that any Socrates
doesn’t qualify as an “ultimate constituent of reality” on any reasonable delineation of
that term. Socrates’ existence supervenes on that of innumerable psychological and bio-
logical events. In their turn, those events supervene on innumerable cellular events
which, in their turn, supervene on innumerable molecular events. And so on – until we
reach the realm of the sub-atomic realm, where there are no objects, in any ordinary
sense of the word, but only displacements of minute quanta of mass-energy.16
Of course, there are possible worlds where Socrates’ existence supervenes on those
of events that are numerically, and even to some extent qualitatively, distinct from
those on which his existence supervenes this world. But there are surely no worlds of
which Socrates is a fundamental constituent, even after we take into account the vague-
ness of that term.17 Consider the properties that are individuative of Socrates. He falls
into a certain biological category. In connection with this, he originated from certain
biological events – on a certain occasion, a certain man and a certain woman created a
certain zygote that developed by way of certain biological processes into an entity hav-
ing certain biological and psychological characteristics. Anything remotely worthy of
being classified as a fundamental constituent of the space-time manifold wouldn’t
come close to being able to participate, in the relevant manner, in biological processes
like those just described.
Of course, what we just about Socrates is true (mutatis mutandis) of any biological
entity. It is not hard to see how that line of thought can be extended to apply to inani-
mate objects. Consider an arbitrary rock. Now consider the processes which brought
it into existence – the confluence of micro-events constituting that rock. Nothing sim-
ple – nothing comparable to a quark or a muon a fortiori the mass-energy displace-
ments constituting such things – would be capable of participating in that causal se-

16. The point just made – that what is linguistically simple is seldom, if ever, what is actually
simple, and that, consequently, facts about language saddle us with some very wrong views about
logic and metaphysics – is urged by Russell himself in a number of places. For example, see Russell
(1959: 238-245). If I am right, then (ironically) this very insight of Russell’s undermines Russell’s
point that we cannot be acquainted with propositions that have constituents of the external as
components.
17. Merrick (2001) holds that people are ultimate constituents of the spatiotemporal world.
Otherwise, his view is in close agreement with ours. If, like Merrick (and Plato and Bishop But-
ler), you believe that people are fundamental constituents of spatiotemporal reality, then replace
my mention of “Socrates” with mention of an inanimate object.
 Chapter 3.  Uniquely individuating descriptions 

quence in the same way as the rock. In general, the objects in terms of which we are
inclined to think – rocks, trees, vases, and so on – are far from fundamental.
So what is fundamental? Socrates’ existence supervenes on facts about organs,
organelles, molecules, atoms, and so on. But everything that we just said about Socra-
tes is true (mutatis mutandis) of each of those other objects. Ultimately, then, Socrates’
existence must supervene on states of affairs that are not object-involving. The only
states of affairs that satisfy this condition are those that consist of instances of proper-
ties, e.g. instances of a certain kind of charge. (Of course, some properties are such that
any instances of them are ipso facto object-involving, e.g. the property of being identi-
cal with Socrates. In this context, then, when we talk about “properties”, we are consid-
ering only those that are not object-involving, i.e. we are considering only those prop-
erties P such that it is not the case that, if P is instantiated, some object, in any ordinary
sense of the word, ipso facto exists.) Socrates is a series of instances of properties.
Of course, this is not quite the right way of putting the matter. Let S be the series of
property-instances that, in actuality, realized Socrates. So S comprises every single elec-
tron-jump involved in every atom that at any point constituted Socrates. Obviously there
are possible worlds where Socrates exists but S does not. After all, Socrates didn’t have to
be composed of the exact muon-displacements that, in fact, composed him.
But, given any possible world where Socrates exists, so does some series that is
non-trivially like S. Given Kripke’s (1972) plausible points about the essentiality of
origins, there is no denying that, in any world where Socrates exists, there is a series S*
such that the earlier parts of S* are extremely similar to the earlier parts of S. But a
stronger claim is warranted. If S* is a series that (in some world other than our own)
realizes Socrates’ existence, then not only must the events that initiate S* resemble
those that initiate S: the non-initial parts of S* must, to some non-trivial degree, resem-
ble the non-initial parts of S. For Socrates to exist in possible world W, it obviously isn’t
enough that a certain egg meet a certain sperm at a certain time. It is also necessary
that the resulting gamete be governed by highly specific biological mechanisms. Oth-
erwise it won’t develop into anything that has a brain or any other organs and that,
consequently, could not be plausibly identified with Socrates.
But the story doesn’t end there. The brain-centers that mediate the psychological
characteristics that we associate with human beings are the last to develop. So the em-
bryo that develops out of the gamete that initiates S* must also follow a path not entirely
unlike the one followed by the fetus that, in actuality, was to become Socrates, the same
thing (mutatis mutandis) being true of the embryo that develops out of that fetus.
We must also remember that both S* and S represent the unfolding of some one
genetic program. After all, S* won’t realize Socrates if the genetic material carried by the
gamete that initiates S* isn’t identical, or nearly so, with the genetic material carried by
the gamete that initiates S. If that material is sufficiently different, then S* will, at best,
realize a child who has the same parents as Socrates but isn’t Socrates. So the genetic
program that is driving the biological developments realized by S* is identical with the
one that drives the biological developments realized by S. It is a distinct possibility that
 Conceptual Atomism and the Computational Theory of Mind

the genetic mechanisms that program for the aforementioned biological developments
are not neatly separable from the genetic mechanisms that program for the generation
of the brain-structures that distinguish Socrates from other human beings. If this is the
case, then supposing that S* progresses in such a way as to guarantee the generation of
a being having any of the psychological properties peculiar to human beings, S* will
ipso facto develop into a person that (even when compared with other people, let alone
non-human objects) bears a non-trivial degree of resemblance to the actual Socrates. In
other words, if S* produces the physiological structures sufficient for any kind of hu-
man personality, then S* will also produce a personality that is at least reasonably simi-
lar to that of the actual Socrates, the reason being that the genetic precursors of those
physiological structures are probably not wholly separate from the genetic precursors
of those personality-traits. If this line of thought is not completely wrong-headed, then
Socrates couldn’t have been that different from how he actually was.
One might respond by saying that, if the laws of biology were sufficiently different,
then S* might realize the events needed for Socrates’ existence without also realizing
any of the brain-structures that distinguish Socrates from others. The appropriate
counter-response is to say that Socrates couldn’t possibly exist in a world that was not
biologically identical with ours. Just as the chemical types found in a world couldn’t
exist in a world that wasn’t governed by the same chemical laws as ours, so the bio-
logical types found in a world couldn’t exist in a world that wasn’t governed by the
same biological laws as ours.
We know, of course, that Socrates could have been different from how he was: he
could have become a soldier as opposed to a philosopher; he could even have spent all
of his years after the age of ten in a vegetative coma. But he could not have become a
turnip or a cloud. We evaluate other people in terms of questions like the following:
What do they do professionally? Are they in a coma? Have they written any books? Are
they in jail? Relative to those benchmarks, Socrates might indeed have had a life very
different from the life he actually had. But relative to benchmarks that that are less
deeply rooted in our emotional parochialisms, Socrates could not have been all that
different from how he was. In any world where Socrates exists, he is, for the duration of
his existence, an entity that, in global terms, satisfies narrowly defined genotypic and, to
a lesser extent, phenotypic conditions. He is not a philosopher in every such world. But
there is no world where, from a strictly biological standpoint, the difference between
how he is in that world and how he is in this world is more than infinitesimally small
compared with the difference between how he is in this world and how, say, a carrot is
in any world. So we may conclude that, in any world where Socrates exists, his existence
supervenes either on S or on some series S* that, in the big scheme of things, is only
slightly different from S. In any case – and this is the important point in this context – in
any such world, Socrates’ existence supervenes on that of some series consisting, of
states of affairs that, in their turn, consist exclusively of property-instances that (ulti-
mately) are not object-involving.
 Chapter 3.  Uniquely individuating descriptions 

We must make one more point before we can see the epistemological consequenc-
es of this metaphysical discussion. As we saw earlier, given two people x and y who are
molecule for molecule duplicates, and given any purely descriptive proposition P, x
grasps P iff y grasps P. Thus, whether one grasps a purely descriptive proposition su-
pervenes on one’s non-relational properties. A corollary is that, if one person x grasps
some purely descriptive proposition P that some other person y does not grasp, then x
necessarily has causal properties that y lacks.
Now we can close the argument. As we’ve just seen, for Socrates to exist in a world
w is for some purely descriptive proposition P to hold in w. It follows that, contrary to
what Russell maintains, we can, at least in principle, grasp a proposition P such that P
is true in w only if Socrates exists in w.
But it doesn’t follow that the content-externalist is right. The content-externalist
says that we in fact grasp propositions whose truth presupposes Socrates’ existence –
that just such a proposition is grasped by anyone who understands sentences like “Soc-
rates was wise.” But we have found this not to be the case. We’ve seen that to have a
concept of Socrates is to know a proposition (an existence-claim) that describes Socra-
tes. But the purely descriptive propositions in question are never anything like those
described a moment ago; they are never propositions that describe the sequence of
property-instances on which Socrates’ existence supervenes. Indeed, although they in
fact have the property of being uniquely satisfied by Socrates, those propositions never,
or at least seldom, necessarily have that property.
There is another reason why the content-externalist doesn’t benefit from our rejec-
tion of Russell’s claim that we cannot be acquainted with external objects. Let P be any
proposition such that P is true in a world w only if Socrates exists in w. The externalist
holds that we can grasp P. That, we found, is actually correct, and Russell was wrong to
deny it. But notice that P doesn’t presuppose Socrates’ existence. Even though P is true
only in worlds where Socrates exists, P exists (though it is false) in worlds where Socrates
does not. P affirms the existence of conditions that are sufficient for Socrates’ existence
and it therefore doesn’t presuppose that such conditions obtain. For that reason, P exists
and can (in principle) be grasped in Socrates-free worlds. P is therefore not object-in-
volving with respect to Socrates, and it is thus not the kind of proposition that, according
to the content-externalist, one grasps when one thinks about, or sense-perceives, Socra-
tes. In fact, if the argument just given is cogent, no propositions are object-involving with
respect to Socrates or, by obvious extensions of that argument, any other external object.
 Conceptual Atomism and the Computational Theory of Mind

All propositions that appear to be object-involving with respect to external objects turn
out to be purely descriptive propositions in disguise.18

18. It occurs to me that this last statement entails that Russell was right to hold that no proposi-
tions that we grasp are object-involving with respect to any constituents of the external world. I
was therefore wrong to say that Russell was wrong. Since this book is going to press in a few days,
I don’t have time to make the changes to this chapter that are needed to make it coherent. In any
case, that incoherence is without significance, since it in no way undermines our main point,
namely: although we grasp propositions that describe constituents of the external world, we don’t
grasp any that have such constituents as veritable components. It is a matter of secondary impor-
tance that I wrongly took Russell’s affirmation of this important truth to be incorrect.
 chapter 4

Some semantic consequences or our analysis

Tokens versus types, semantics versus pre-semantics

We have seen that what is literally meant by an utterance must be distinguished from the
information that is uploaded or absorbed in the processing of computing its literal mean-
ing. There is a closely related distinction that we have thus far only briefly discussed: the
distinction between expression-types and expression-tokens. Given a proper under-
standing of this distinction, it will be possible to refine and extend our findings.

Expression-types versus expression-tokens

The expression-type “you are tall” isn’t true or false. It is different tokens of it that are
true or false. Where sentences containing indexicals are concerned, it is de rigueur to
distinguish between type-meaning and token-meaning. (An “indexical” is a system-
atically context-sensitive expression, e.g. “I”, “you”, “this”, “that”, “here”, “now.”)
It follows that the distinction between sentence-types and sentence-tokens must
be made universally. An utterance of “Smith is in the kitchen” at t may be true while
such an utterance at t* is false. This means that those utterances encode different prop-
ositions. But how his is this possible? After all, “Smith is in the kitchen” appears not to
contain an indexical component.
In actuality, that sentence does have an indexical component. That sentence is
tensed, and because of the tense-marker what a token of it expresses is a function, in
part, of the time of tokening. If uttered at different times, tokens of the sentence-type
“Smith is in the kitchen” express different propositions. Hence, the tense-marker is an
indexical that picks out times. For this reason, the content of that sentence-type cannot
be identified with any one proposition. And it is not plausible to say of that type that it is
ambiguous between infinitely many different propositions of the form Smith is in the
kitchen at time t. (“Smith is in the kitchen” is not comparable to “Smith is dumb.” The
latter is ambiguous; the former is not.) Given these points, what we must say is that the
semantic content of that sentence-type is not a proposition, and is instead a function that
assigns propositions to tokens of that type, depending on when those tokens occur. So,
while tokens of that sentence-type have propositions for their semantic contents, the
type per se does not. Given any sentence-type containing a tensed verb – and this means
any sentence-type belonging to natural language – the semantic content of that type is
not a proposition, but is rather a function that assigns propositions to its tokens.
 Conceptual Atomism and the Computational Theory of Mind

Of course, in some cases, tense is obviously irrelevant. The tense-marker in a token

of “two plus two equals four” is idle. But this is easily explained along pragmatic lines:
the tense-marker may be ignored, given the obvious mathematical (not semantic) fact
that temporal considerations are irrelevant to the truth-conditions of arithmetical
statements. It would be arbitrary, and also unnecessary, to say that the tense-marker in
“two plus two equals four” had a different semantics from the tense-marker in “John
swims every day.”
In artificial extensions of natural language, one may encounter tenseless sentences.
At least arguably, a written token of “2+2=4” has no tense. But considerations of uni-
formity demand that we see the corresponding expression-type as encoding, not a prop-
osition, but a function that assigns a proposition to tokens of that type. Sentence-types
are platonic entities; sentence-tokens are spatiotemporal. We’ve seen that, where natural
language is concerned, it is sentence-tokens, not sentence-types, that bear propositions.
Given this, suppose we said that the expression-type “2+2=4” had a proposition for its
semantic content. In that case, we would be saying that the property of being proposi-
tion-bearing was sometimes a property of the spatiotemporal and was sometimes a
property of the platonic. But such a proposal would bifurcate the notion of literal mean-
ing, given how deep the division is between the platonic and the spatiotemporal.
We must also remember that a language is a means of communication. The things
that are heard and spoken, read and written, are tokens. Those things can only be un-
derstood on the basis of a knowledge of the relevant semantic rules. Therefore, those
rules must assign propositions to sentence-tokens. This suggests that, so far as sen-
tence-types have a semantic function, it is only to assist in the assigning of proposi-
tions to sentence-tokens. And this, in turn, confirms our view that the semantic con-
tents of sentence-types are functions of the kind earlier described.

Pre-semantic meaning as coinciding with type-meaning

Before moving on, let us discuss why the distinction between sentence-types and sen-
tence-tokens is relevant to a work on conception.
Let T be a token of “that boorish uncultured man over there (the one who’s been
dominating the conversation with his unfunny jokes for the last hour) is a spy”, and
suppose that Smith is the man in question. Let T* be a token of “that erudite and illu-
minating gentleman next to the fireplace (the one who has been discoursing in such a
brilliant and yet unassuming fashion for the last ten minutes) is a spy.” As before, sup-
pose that Smith is the man in question.
As we’ve noted, Kaplan (1989) argued compellingly that T and T* encode the very
same proposition, namely Smith is a spy. So there is some individual x (namely Smith)
such that, at the level of semantics, each of T and T* encodes a proposition that is true
just in case x is tall. At first it seems as though Kaplan’s analysis must be wrong, given
 Chapter 4.  Some semantic consequences or our analysis 

that T and T* communicate different, and indeed non-equivalent, propositions. But in

Chapter 1, we showed why Kaplan’s analysis is perfectly consistent with this fact.
Notice that our explanation made heavy use of the distinction between type-se-
mantics and token-semantics. One assigns the right meaning to T and T* on the basis
of a knowledge of the relevant expression-types. So one works through type-meaning
to ascertain token-meaning. As a result, what is communicated by a sentence-token is
a function, at least in part, of type-meaning. A consequence is that two tokens that
have the very same proposition for their literal meanings may communicate very dif-
ferent propositions.
As we will soon see, what we just said about sentence-tokens containing demon-
strative-expressions is true of sentence-tokens containing definite descriptions and,
indeed, of all sentence-tokens. If we keep in mind the distinction between type-mean-
ing and token-meaning, a number of otherwise unexplainable facts fall into place.
And, as we will now see, if we don’t keep that distinction in mind, we are forced to
adopt some counterintuitive and even incoherent views.

Substitution-failures

The sentence:
(1) “Bill Gates is smart”
is what results when the definite description in:
(2) “the richest man in America is smart”
is replaced with a co-referring expression. Even though (1) and (2) have the same
truth-value, they seem to have different meanings. So it certainly appears that substi-
tuting a referring term with a co-referring term can change meaning. It also appears
that such intersubstitutions can sometimes change truth-value. Given reasonable as-
sumptions about what Smith believes,
(s1) “Smith thinks that Bill Gates is smart”
and
(s2) “Smith thinks that the richest man in America is smart”
may certainly appear to differ in truth-value, and not only in meaning.
A related point is that:
(3) “Bill Gates is Bill Gates”
is trivial, and therefore necessarily true, whereas
(4) “Bill Gates is the richest man in America”
 Conceptual Atomism and the Computational Theory of Mind

is non-trivial and also contingent.

Much of what is generally believed about substitutivity is in need of serious revi-
sion, given what we have said thus far in this work.
As we’ve seen, a token T of
(5) “you are tall”
may have the same literal meaning as a token T* of
(6) “that guy (the one standing over there, reading War and Peace while leaning
against a statue of a dancing walrus) is tall”,
even though T and T* communicate dramatically different propositions. (For the pur-
poses of discussion, let us henceforth assume that the occurrence of “you” in (5), and
in any expression containing (5), co-refers with the occurrence of “that guy” in (6), and
in any expression containing (6).) We have already explained why there is no difficulty
explaining this gulf between literal and communicated meaning. What we said about
T and T* is easily extended to show that, in general, intersubstituting co-referring
terms cannot change literal meaning, even though doing so often has dramatic effects
on communicated meaning. We will find, in other words, that there are no intensional
contexts. We will see that a failure to distinguish semantics from pre-semantics, along
with a subsequent failure to distinguish reference from quantification, underlies the
spurious but widespread view that there are intensional contexts.
Notice that there could certainly appear to be a difference in truth-value between
a token sT of:
(s5) “Smith thinks that you are tall”
and a token sT* of:
(s6) “Smith thinks that guy (the one standing over there, reading War and Peace
while leaning against a statue of a dancing walrus) is tall.”
Obviously the apparent difference in truth-value between sT and sT* correlates with the
apparent difference in meaning between T and T*. But the latter difference is merely
apparent, as we saw. The same is therefore true of the former. For some x, each of sT and
sT* encodes the bare singular proposition: Smith thinks that x is tall. In any case, given
the points made in connection with the apparent (but non-actual) differences in mean-
ing between T and T*, there is no difficulty explaining why what is communicated by sT
may differ in truth-value from what is communicated by sT*.
When we take into account the information through which one must compute the
literal meanings of (sT) and sT*, it becomes clear why those tokens would inevitably
convey very different propositions – it becomes clear why one of them would convey or
suggest that the addressee knew of the existence of a dancing-walrus sculpture, while
the other would not. Thus, once we take pre-semantics into account, there is no need to
assume any difference in literal meaning, or therefore in truth-value, between sT and sT*.
 Chapter 4.  Some semantic consequences or our analysis 

Further, if we were to posit such a difference, in order to explain the apparent differ-
ences in truth-value between those tokens, our explanation would be superfluous. In-
deed, it would be in conflict with an incompatible, but adequate, explanation.
What we see, then, is that many apparent substitution-failures involve no change in
literal meaning (or, therefore, truth-value), and are to be understood pragmatically. More
specifically, they are to be understood in terms of pre-semantics. Let us extend now this
line of thought by considering the controversial topic of definite descriptions.
Once again consider the sentence:
(2) “the richest man in America is smart.”
(2) certainly seems to have a very different meaning from:
(7) “the first person to make a fortune by selling personal computers is smart.”
But, it would seem, (7) is what results when a referring term in (2) is replaced with a
co-referring term. Similarly,
(s2) “Smith thinks that the richest man in America is smart”
seems to have a different truth-value from:
(s7) “Smith thinks that the first person to make a fortune by selling personal com-
puters is smart.”
In any case, (s2) and (s7) appear to have very different truth-values, given certain rea-
sonable assumptions about Smith’s beliefs, e.g. that Smith believes that, despite his
stupidity, Jones was the first person to make a fortune from selling personal computers
and, moreover, that Bill Gates is smart.
Frege and Russell were keenly aware of the apparent differences in meaning just
discussed, and they took them to be actual differences in meaning. Each explained
these putative differences in different ways. Both explanations, we will see, have radi-
cally counterintuitive consequences (especially Frege’s). Further, neither explanation is
necessary. The problems that they are meant to solve can be solved without advocating
theories that have any of the unpalatable consequences of their theories. Let us start
with Russell’s theory.

Russell’s Theory of Descriptions

What is conveyed by (2) is different from what is conveyed by

(1) “Bill Gates is smart”,
and also from what is conveyed by:
(7) “the first person to make a fortune by selling personal computers is smart.”
 Conceptual Atomism and the Computational Theory of Mind

And, of course, (1) and (7) differ from each other in respect of what they convey. Let us
now try to state more specifically what each of these sentences conveys.
The proposition conveyed by (2) obviously involves the concept richest man in
America. The word “the” connotes existence. (2) is not true if nobody is a richest man
in America. (Suppose that there were no men at all in America. In that case, obviously,
(2) wouldn’t be true. For parallel reasons, (2) wouldn’t be true if there were no richest
man in America.) Further, unlike the words “four” and “some”, the word “the” con-
notes uniqueness. If there are a hundred equally rich richest men in America, then (2)
fails to be true. So a necessary condition for the truth of (2) is that there be somebody
who is uniquely a richest man in America. Of course, if that person is not smart, then
(2) is false; and if that person is smart, then (2) is true. So (2) is true exactly if the fol-
lowing proposition is true:
(2P) There is somebody x such that x is uniquely a richest man in America; moreo-
ver, x is smart.
For exactly similar reasons, the proposition meant by (7) is identical, or at east equiva-
lent, with:
(7P) Somebody x is uniquely such that x was a first person to make a fortune sell-
ing personal computers; moreover, x is smart.
In general, Fthe phi has psiG is true exactly if the proposition something x uniquely has
phi; moreover, x has psi is true. Thus, “the richest man in America” doesn’t refer to Bill
Gates. It doesn’t refer to anything. Rather, as Kent Bach put it, it “contributes its quan-
tificational structure” to the meanings of the sentences (or, more accurately, the sen-
tence-tokens) in which it occurs. It is thus a quantifier, not a singular term.1

Evaluating the Theory of Descriptions

The relevant data can be accounted for without accepting the dubious view that “the
inventor of bifocals” doesn’t refer to the inventor of bifocals. For reasons already given,
we must distinguish types from tokens. Given this, suppose that the general semantic
rule for definite descriptions were as follows:
(*) a token T, that occurs at time t, of Fthe phiG is a singular term that refers to x
exactly if, at t, x is a unique phi; and if there is no such individual at t, then T
is semantically empty.
Suppose, therefore, that if x is a unique phi at a given time, then a token at that time of
Fthe phiG has x for its sole semantic content and, consequently, that no concept of x –

1. Bach (1984).
 Chapter 4.  Some semantic consequences or our analysis 

no Fregean sense or Russellian description – is any part of what is literally meant by

such a token.
If (*) is right, then a token of any sentence of the form Fthe phi has psiG would ei-
ther have the bare singular proposition x has psi (for the appropriate value of x) for its
literal meaning, or it would fail to encode a proposition (i.e. it would fail altogether to
have a truth-evaluable literal meaning). It would mean x has psi if there were, at the
time of utterance, a unique phi; and it would have no proposition for its literal mean-
ing if there were no such individual. At the same time, for the reasons given earlier,
such a token would convey the existence-claim: something x uniquely has phi; moreover,
x has psi. So such a token would convey exactly the proposition that, according to Rus-
sell, is its literal meaning. Thus, the supposition that definite descriptions are singular,
directly referential expressions is entirely consistent with the facts about cognitive sig-
nificance that motivate Russell’s theory. In fact, that supposition explains those very
facts, given the distinction between semantics and pre-semantics.2
If (*) gives the general semantic rule for definite descriptions, then the semantics
of “the richest man in America” will be given by the rule:
(**) a token T, that occurs at time t, of “the richest man in America” is a singular
term that refers to x exactly if, at t, x is a unique richest man in America; and
if there is no such individual, then T is semantically empty.
Supposing that (**) is the correct rule, a token in 2006 of (2) has the bare singular
proposition Bill Gates is smart for its semantic content, given that in 2006 Bill Gates is
the richest man in America. But for the reasons we’ve discussed, such a token would
still communicate the existential proposition:
(2p) Somebody x is uniquely a richest man in America (at the present time); more-
over, x is smart.
For the sake of argument, let us suppose that occurrences of “the richest man in Amer-
ica” are singular terms. (This is what our pre-theoretic intuition strongly suggests.) The
semantic rule for “the richest man in America” is obviously not going to be:
(***) “The richest man in America” refers to Bill Gates.
The semantic rule for that expression must be general; it must not presuppose that this
or that particular individual is the richest man in America or even whether there is
such a man. So that rule will necessarily be at least approximately like:
If, at time t, there is a unique individual x who is a richest man in America, then
an occurrence at t of “the richest man in America” refers to x.

2. For reasons that are not important in this context, (*) is only an approximately statement of
the general rule for semantic descriptions. A less approximate statement of that rule is this:
(*C) Given any context C, if there is exactly one salient thing x in C which has phi, then a
token t of Fthe phiG in C has x for its sole semantic content.
 Conceptual Atomism and the Computational Theory of Mind

Of course, the rule just stated is a variant of (**). Anyone who has the linguistic knowl-
edge needed to figure out what is meant by a token of (7), or by any other sentence-
token containing an occurrence of “the richest man in America”, will know that a token
of that expression refers to an object x just in case there is a unique richest man in
America, and x is that person. (Of course, this knowledge will be subpersonal, like a
non-linguist’s knowledge of the rules governing when to say “me” as opposed to “I.”3
My analysis does therefore assume that there is subpersonal knowledge. We will dis-
charge that assumption in Chapter 18. In any case, there is a vast amount of experi-
mental and theoretical support for that assumption, granting that many philosophers
are virulently opposed to it.) So anyone who can understand a token t of (7) will know
that t is truth-evaluable (i.e. either true or false) only if somebody x is, at the time of
tokening, a unique richest man in America. For similar reasons, such a person will
know that t is true only if x is smart. So that person will know (albeit implicitly or
subpersonally) that t is true exactly if, at the time of tokening, somebody x is a unique
richest man in America; moreover, x is smart. (Suppose that a competent speaker of
English who knows that there is no king of France hears the sentence “the king of
France is bald.” As Strawson (1951) pointed out, that person will have a strong inclina-
tion to regard that sentence-token as abortive, i.e. as neither true nor false. Contrary to
what Russell’s theory would seem to entail, and to what Russell himself (1905) explic-
itly says, one would be guilty of deviant linguistic behavior were one to respond by
saying: “that is false; for there is no king of France.” I propose that our pre-theoretic
inclination to regard sentence-tokens like “the king of France is bald” as abortive (giv-
en that there is no king of France) expresses our (implicit or subpersonal) knowledge
that such sentences presuppose, and thus do not assert, that there is a king of France.
In any case, Russell’s theory has no explanation for the fact that we instinctively regard
as neither true nor false sentence-tokens that contain empty definite descriptions,
whereas the theory that I am advocating – according to which such sentence-tokens
really are neither true nor false – obviously does have an explanation of that same fact.
Russellians try to neutralize problems of this kind by saying that it is pragmatics, not
semantics, that is to blame for the apparent deviation between their theory and the
linguistic data. But, as we will soon see, pragmatics cannot do what the Russellian
needs it to do.)
Here is another way of thinking about the matter. For the sake of argument, let us
continue to suppose that occurrences of definite descriptions are singular terms, as
opposed to quantifiers. In that case, the semantic rule for (7) is not going to be:
(****) “The richest man in America is smart” means: Bill Gates is smart.

3. Of course, even a linguist’s knowledge of these rules is subpersonal. That is, the knowledge
that enables the linguist to speak grammatically is subpersonal. In his capacity as a linguist, he
may generate analogous knowledge at the personal level. But we are dealing with two different
representations of the same (or analogous) information. We are not dealing with one instance of
knowing a given set of rules, but with two such instances (that happen to be in the same mind).
 Chapter 4.  Some semantic consequences or our analysis 

Rather, it is going to be:

If somebody x is uniquely a richest man in America at a given time, then a token t (at
that time) of “the richest man in America is smart” means (or is true iff) x is smart.

Anyone who understands such a token works through that rule. So such a person
knows that, in order for a token of that sentence to be correct, there must (at the time
of tokening) be somebody x such that x is uniquely a richest man in America and such
that x is smart.
So to explain why occurrences of (2) and (7) have the cognitive significances that
they in fact have, it isn’t necessary to deny that occurrences of “the richest man in
America” and other definite descriptions are singular terms; and it therefore isn’t nec-
essary to suppose that such occurrences are quantifiers. So far as it seems necessary to
take such a view, that is because one is operating within a one-dimensional semantic
system and is thus operating on the false assumption that (***) and (****) are the rel-
evant semantic rules.4
To sum up, given the distinction between semantics and pre-semantics, and be-
tween token- and type-meaning, the thesis that occurrences of definite descriptions
are singular terms is easily reconciled with the facts concerning what is conveyed by
sentences containing such occurrences.
In light of these points, we can easily reconcile the supposition that definite de-
scriptions are singular terms with the apparent difference in truth-value between:
(s2) Smith thinks that the richest man in America is smart
and
(s7) Smith thinks that the first person to make a fortune by selling personal com-
puters is smart.
For reasons that we’ve already covered, even if it is assumed that definite descriptions
are singular terms, tokens of the underlined expressions will communicate Russell-
style existence-claims. More exactly, such tokens will communicate
(2P) somebody x is uniquely a richest man in America; moreover, x is smart.
and
(7P) Somebody x is uniquely a first person to make a fortune selling personal com-
puters; moreover, x is smart.

4. See Kuczynski (2006d).

 Conceptual Atomism and the Computational Theory of Mind

It is therefore to be expected that tokens of s2 and s7 would communicate (but not se-
mantically encode):
(s2P) Smith thinks that somebody x is uniquely a richest man in America and,
moreover, that x is smart.
and
(s7P) Smith thinks that somebody x is uniquely a first person to make a fortune sell-
ing personal computers and, moreover, that x is smart.
Of course, (s2P) and s7P could well have different truth-values, explaining why intersub-
stituting co-referring definite descriptions may change the truth-value of what is com-
municated (though not of what is literally meant).
There is a nuance that we must take into account if our analysis is to be precise. It
is well known that s2 and s7 have both “wide-scope” and “narrow-scope” readings.5 s2
can convey that, without knowing who the richest man in America is or even whether
there is such a man, Smith knows of that person that he is smart. This would hold in a
scenario like the following. Smith lives next to Bill Gates. Smith knows that Bill Gates
is smart, but not that Bill Gates is particularly wealthy. This is the wide-scope (de re)
reading, and it is given by the proposition:
(s2ws) Somebody x is uniquely a richest man in America, and Smith thinks that x is
smart.
Given that occurrences of 2 convey 2P, even on the assumption that tokens of definite
descriptions are directly referential, it is no wonder that s2 should be capable of con-
veying (s2ws). There is no reason why the occurrences of “Smith thinks that” in tokens
of s2 shouldn’t be capable of having different degrees of scope – no reason not to think
that, for reasons of pragmatics, those occurrences could be given either wide-scope or
narrow-scope, depending on the specifics of the context of tokening. Indeed, it would
be artificial to suppose that contextual and pragmatic factors did not have such an ef-
fect. So our analysis explains why occurrences of s2 can communicate (S2ws), without
having that for their literal meaning. (Of course, everything that we are saying about s2
is true mutatis mutandis of s7.)
The “narrow-scope” reading of s2 is given by s2P. We’ve already explained why to-
kens of s2 can convey (s2P).
There is thus no need to advocate Russell’s theory to explain the relevant facts
about cognitive significance, i.e. about what is conveyed by sentence-occurrences of
the form F…the phi…G This completely undercuts Russell’s theory, at least so far as that

5. See Russell (1905) and Neale (1990).

 Chapter 4.  Some semantic consequences or our analysis 

theory’s raison d’être is that it accounts for what is communicated by sentence-tokens

containing definite descriptions.6
But not only is Russell’s theory unnecessary. It is demonstrably false. We must begin
by distinguishing between semantic and conceptual analysis. A semantic analysis of an
expression identifies its literal meaning. A conceptual analysis gives an analysis of that
proposition itself, i.e. it identifies a proposition that is logically equivalent with the one
that is literally meant, but is more perspicuous in respect of its inferential structure.7
Given this distinction, there are two possible interpretations of Russell’s theory.
We can say that he is doing semantic analysis (identifying literal meanings) or that he
is doing conceptual analysis.
Let us consider both interpretations, starting with the first. Consider the thesis
that (2) is synonymous with (2P). For two sentences to be synonymous is, by definition,
for them to have exactly the same literal meaning. It is not easy to find two distinct
sentences , belonging to a single language, that have precisely the same literal meaning.
But “Smith is an enemy” and “Smith is a foe” are synonymous (or close enough for our
purposes). Therefore a Southerner’s utterance of “Smith is an enemy” will be synony-
mous with a Northerner’s utterance of “Smith is a foe”, it being irrelevant that a North-
erner’s accent is different from a Southerner’s. If two utterances are synonymous, the
differences between them are innocuous (except in respects that are plainly irrelevant
in the present context).
But the differences between Fthe phi has psiG and its Russellian paraphrase are not
innocuous. Suppose that somebody asks you the following question:
(Q) Is it true that there is a unique richest man in America? Or are there in fact
several equally rich richest people in America?

If you answer by saying:

(2) “the richest man in America is smart”,
your answer displays a lack of linguistic competence. Granting that (2) conveys an answer
to the question, no one who speaks English would think that (2) was a linguistically cor-
rect answer. And the reason is obviously that you have presupposed the very thing that,
under the circumstances, you were supposed to assert. A linguistically correct (if rather
pedantic) answer would be given by an utterance of “there is a unique richest person in
America” or “somebody x is a unique richest person in America” or, therefore:
(2P) “somebody x is a unique richest person in America; moreover, x is smart.”

6. There are other reasons for advocating Russell’s theory. These relate to issues having to do
with variable-binding and anaphora. See Neale (1990). For reasons that I have given elsewhere
(Kuczynski 2005c), I believe that these reasons for holding Russell’s theory are spurious. But a
discussion of this would take us off our present course.
7. See Kuczynski (2004b, 2006d).
 Conceptual Atomism and the Computational Theory of Mind

In general, it is clear that, in a context where an utterance of Fthe phi has psiG would
show linguistic incompetence, its Russellian paraphrase (Fsomething x uniquely has
phi; moreover x has psiG) would show linguistic competence. And, of course, the same
argument mutatis mutandis shows that, in a context where an utterance of Fthe phi has
psiG would show linguistic competence, its Russellian paraphrase (Fsomething x
uniquely has phi; moreover, x has psiG) would show linguistic incompetence.
If (2) and (2P) were synonymous, then the differences between them would be as
innocuous as the differences between a Southerner’s and a Northerner’s pronuncia-
tions of a given sentence. Alternatively, those differences would be as innocuous as the
difference between “Smith is an enemy” and “Smith is a foe.” But (2) and (2P) are very
much not interchangeable. In fact, it would be hard to think of a single context where
the one would be appropriate if the other were. This cannot be chalked up to pragmat-
ics since their non-interchangeability is rooted in facts about linguistic competence,
and not in facts about sensitivity (or a lack thereof) to situational nuances.8
A story may clarify this last point. You say to an employee: “your performance
here has been less than exemplary; and I think it would be a good idea if you started
clearing your desk.” Of course, what you are conveying is: you’re fired. But the em-
ployee is so unintelligent that he takes your words literally. He responds to your state-
ment by saying (he is not attempting to be ironic or humorous):
“You’re right, boss. I probably should clear my desk – after all it’s pretty dirty. And,
just as you say, my performance here has been bad. Thanks for telling me. I will
now clean my desk and attempt to improve my performance.”

The employee has shown a kind of incompetence. But it is not linguistic incompetence.
He understood the literal meanings of your words perfectly well. The employee has
shown a kind of emotional tone-deafness. What is needed is the intervention of a psy-
chotherapist, not of a speech-pathologist. But somebody who says (2) where (2P) is ap-
propriate, or vice versa, is displaying linguistic incompetence. They haven’t mastered the
relevant semantic rules for those expressions. There are many contexts where uttering
the one sentence would be linguistically inappropriate – where it would show that the
speaker needed to take more English-classes (supposing that he were not intentionally
flouting the rules of English semantics) – and where uttering the other would not.
Linguistic deficiencies are non-pragmatic. Pragmatics is the interface of literal
meaning and context. A “pragmatic deficiency”, such as that evinced by the employee,
does not show a failure to have internalized the relevant literal meanings, but instead
shows a failure to see how those meanings interact with facts about culture, psychol-
ogy, and the like. Somebody who says Fsomebody x is a unique phi; moreover, x has
psiG where the right statement is Fthe phi has psiG, or vice versa, is showing grammati-
cal-semantic incompetence.

8. See Mates (1952) for a similar line of thought.

 Chapter 4.  Some semantic consequences or our analysis 

The Theory of Descriptions is decidedly implausible when it comes to questions

and imperatives. If Russell’s theory is right, then:
(8) “tell the guy over there to call the fire-department”
means either:
(8NS) Make it be the case that somebody x is uniquely a guy over there; moreover,
tell x to call the fire-department
or
(8WS) There is somebody x such that x a unique person over there, and tell x to call
the local fire-department.
Actually, if Russell’s theory is right, then (8) is multiply ambiguous, the reason being
that it contains two definite descriptions. One of the remaining disambiguations is:
(8DNS) Make it be the case that there is exactly one local fire-department x, and make it
be the case that there is exactly one guy over there y; and tell y to call x.
If we insist on eliminating all of the definite descriptions in (8), there are three other
disambiguations of that sentence. But for our purposes, we can focus on (8WS) and (8NS).
The problems for Russell’s theory that we uncover in connection with these two sen-
tences apply with extra force to more strict Russellian paraphrases of (8), such (8DNS).
First of all, there is no evidence (independent of Russell’s theory) that (8) is at all
ambiguous – let alone ambiguous between four different meanings. But Russell’s theo-
ry is highly questionable even if we set this aside. For as Searle (1979) observed, it is
hard to believe that, on any disambiguation, an utterance of (8) expresses an order that
you make there be somebody standing in a certain place. Further, there are obviously
many contexts where an utterance of (8NS) would show a dearth of linguistic compe-
tence and where an utterance of (8) would show no such dearth.
For similar reasons, it is hard to believe that, on any disambiguation, (8) is syn-
onymous with (8WS). In uttering (8), you certainly don’t seem to be asserting that there
is a unique man in a certain location. You seem to be presupposing it, not affirming it,
as Strawson (1951) observed. This is confirmed by the fact that an utterance of (8WS)
would show a dearth of linguistic competence in contexts where an utterance of (8)
would show no such dearth. It is therefore hard to believe that (8) is under any circum-
stances synonymous with (8WS).
Advocates of Russell’s theory argue that, so far as (8) strikes us as not being syn-
onymous with (8NS) and (8WS), it is because we are mistaking pragmatic epiphenomena
for literal meaning. They argue that, after we root out the contextual factors that medi-
 Conceptual Atomism and the Computational Theory of Mind

ate between literal and understood meaning, Russell’s theory turns out to be in perfect
accordance with the data.9
But given that synonymous sentences ipso facto have precisely the same meanings,
it is hard to see how replacing (8) with either (8NS) or (8WS) could be so far from in-
nocuous; it is hard to see how doing so could turn a reasonable sounding utterance
into an extremely strange sounding one, or how it could turn an instance of linguistic
competence into one of linguistic incompetence. It is therefore hard to see how replac-
ing (8) with either of those other sentences is comparable to replacing “enemy” with
“foe” or “heather” with “gorse.”
To say that two expressions are synonymous is to make a very strong claim. Obvi-
ously intersubstituting synonymous expressions has phonetic consequences. (Tokens
of “foe” sound different from tokens of “enemy.”) But it is hard to see how such differ-
ences could be semantically pregnant enough to sustain Russell’s theory in the face of
the points that we’ve made. So it is not an option to suppose that Russell’s theory gives
the literal meanings of sentences containing definite descriptions.10
Also, we have seen that the negation of Russell’s theory is favored by a careful scru-
tiny of the contextual factors that mediate between literal and understood meaning,
directly contradicting what Russellians maintain.
What about the idea that Russell’s theory gives, not literal meanings, but concep-
tual analyses, and thus logical equivalences? That idea is not tenable, as the following
argument makes clear.
If the literal meaning of “the richest man in America is smart” is not some quanti-
fied generalization, then “the richest man in America” is not a quantifier. If that expres-
sion is not a quantifier, then what could it be? The only reasonable hypotheses as to the
semantics of that expression are, first, that it is a singular term and, second, that it is a
Russell-style quantifier. So if “the richest man in America is smart” does not have a
quantified generalization for its literal meaning, then “the richest man in America” is a
singular term. If “the richest man in America” is a singular term, then (supposing that
there is in fact a unique richest man in America) there is some O such that a token of
“the richest man in America is smart” means O is smart. But O is smart is not logically
equivalent with something x is uniquely a richest man in America, and x is smart. So Rus-
sell’s theory is simply wrong if it is taken to identify propositions equivalent, but not
identical, with those literally meant by sentences containing definite descriptions.
Before we move on, we must note a fact about the argument just given. That argu-
ment presupposes that, if “the richest man in America” refers to O, then “the richest
man in America is smart” encodes the singular proposition: O is smart. Put another
way, that argument presupposes that all reference is direct reference, where the term

9. See Blackburn (1984: 308-310) and Neale (1990). In Kuczynski (2004) I discuss the viabi-
lity of attempts to blame pragmatics for the apparent discrepancies between Russell’s theory and
the data.
10. See Kuczynski (2004b, 2006d).
 Chapter 4.  Some semantic consequences or our analysis 

“direct reference” is to be defined thus: E refers directly to O just in case, in virtue of

having the form F…E…G, a sentence S encodes a proposition that has O itself - not
some Fregean sense or concept that O falls under - as a constituent. But many seman-
ticists hold that there is also indirect reference, where the term “indirect reference” is
to be defined thus: E refers indirectly to O just in case, in virtue of having the form
F…E…G, a sentence S encodes a proposition that has some Fregean sense that O falls
under - and not O itself – as a constituent. But so-called indirect reference isn’t refer-
ence at all, and is really just quantification. We will see why this is so when, in a mo-
ment, we discuss Frege’s views about reference. Right now, let us state our second rea-
son to believe that Russell’s theory must be seen as an attempt to give literal meanings,
as opposed to mere logical consequences. .
For the sake of argument, suppose that (2)’s literal meaning is logically equivalent
with some quantified generalization. In that case, presumably, that literal meaning is
itself given by a quantified generalization. But if that is the case, then Russell’s theory
does aspire to give literal meanings. But we’ve already seen why that theory does not
give literal meanings.11
Let us develop the argument just given. If (2)’s literal meaning is logically equiva-
lent with some quantified generalization, then (2) is identical in meaning with either:
(i) “something x is such that anything y is a richest man in America just in case
y=x; moreover, x is smart”
or with:
(ii) “for some x, given any y, y is a richest man in America iff y=x; moreover, x is
smart”
or with:
(iii) “some object x has the property of being uniquely a richest man in America;
moreover, for anything y, if y=x then y is smart”
or with some other quantified generalization that is logically equivalent with each of (i)-(iii).
But what we said about (2P) is true of each of these other sentences. Consider (ii), for exam-
ple. Replacing an occurrence of (ii) with an occurrence of (2) would typically produce a
linguistically inappropriate utterance where there was previously a linguistically appropri-
ate one. The same is true of (i) and (iii) and any other sentence that expresses a quantified
generalization.12

11. See Kuczynski (2004b, 2006d).

12. There is another problem with Russell’s theory: it makes it more difficult than it would
otherwise be to explain the obvious grammatical similarities between definite descriptions, on
the one hand, and indexicals and proper names, on the other.
Of course, Russell was acutely aware of the fact that, if his theory is correct, then the gram-
mar of definite descriptions is very misleading as to what is actually meant by such expressions.
He did not seem troubled by this consequence; and he seemed to hold that, in natural language,
 Conceptual Atomism and the Computational Theory of Mind

Here we must echo a point made a moment ago. Some semanticists hold that, contrary to
what is assumed by the argument just put forth, a sentence may have for its litereal meaning
a proposition that, while being logically equivalent with each of (i)–(iii), is nonetheless not
itself a quantified generalization of any kind. My own view is that this position is incoher-
ent. But it was held by no less a thinker than Gottlob Frege, and thus cannot be rejected
before being considered carefully. Let us now discuss this view of Frege’s, along with the
semantic system of which it is a part.

Frege’s view

Frege (1892) correctly held that definite descriptions are singular terms. Nonetheless,
Frege’s theory is less defensible than Russell’s. Sentences of the form F...the phi...G don’t
mean what Russell says they mean. But we could coherently imagine a language such
that Russell’s theory gave the semantics of that language. (Let L be a language that is just
like English except that, by our stipulation, definite descriptions in L have the semantics
that Russell believed English definite descriptions to have. Obviously there is no a priori
reason why L couldn’t exist.) So Russell’s theory, though false, is not incoherent. But
Frege’s theory is incoherent. There is, we will now see, no possible language such that
Frege’s system gives the semantics of that language.
Frege (1892) rightly believed in what has come to be known as the “principle of
compositionality”13: the meaning of a complex expression is a function of the mean-
ings of its constituents. A specific version of this principle relates to reference: the ref-

there was such a pervasive mismatch between grammatical and logical form that it didn’t re-
dound to the discredit of his theory that it put the grammatical properties of definite descrip-
tions out of alignment with their semantic (and logical) properties.
But it is a very controversial question whether semantic theories can embody such a deep
disregard for grammatical facts. Jackendoff (1989) has argued compellingly that a sine qua non
for a semantic theory’s correctness is that it have a certain consonance with grammatical facts.
My own view is that Jackendoff is right. If we reject Jackendoff ’s view, then facts about grammar
become free-wheels – wheels that turn independently of semantics. But surely grammatical facts
are not entirely without semantic significance. In Kuczynski (2005c), I argue that grammatical
facts are semantic facts, and that it is therefore self-contradictory to suppose that grammar mi-
ght be independent of meaning. The basic idea is that the difference between “Smith sees Jones”
and “Smith, the relation of seeing, Jones” is semantic and is also grammatical and that, conse-
quently, we must suppose grammatical inflections to contribute material that connects the deno-
tations of expressions. Grammatical material is connective semantic material. In Chapter 23,
this idea will be developed, though not as completely as it is in Kuczynski (2005).
13. In fact, Frege seems to have introduced this crucial notion to the study of semantics.
 Chapter 4.  Some semantic consequences or our analysis 

erent of a complex referring term is a function of the referents of its parts. (My view is
that both principles are unqualifiedly correct.14)
Let us start with our paradigm:
(2) “the richest man in America is smart.”
According to Frege, the underlined expression is a singular term.15 At the same time,
Frege also held that (2) is logically equivalent with:
(2P) somebody x is a unique richest person in America; moreover, x is smart.
This raises an important question. If the literal meaning of (2) is logically equivalent
with (2P), why isn’t “the richest man in America” simply a quantifier? If Frege is right
about (2), then (2) says nothing more and nothing less than (2P). In general, if Frege is
right about Fthe phiG, then Fthe phi has psiG is true exactly if phi is uniquely instantiated
and any instance of it has psi. So any sentence of the form Fthe phi has psiG answers the
question Fhow many phi’s are psi’s? G
But in that case, Fthe phiG is a text-book case of a quantifier. “No person is smart”
answers the question “how many people are smart?”, the same being true of “some per-
son is smart”, “three people are smart”, and so on. That is why “some person”, “no per-
son”, and so on, are quantifiers.16 If Frege is right, then “the person over there is smart”
answers that very same question. (It gives the answer: “there is exactly one person over
there: and any such person is smart.”) So it isn’t clear how Frege’s position amounts to

14. I actually think that they are truisms or tautologies. I have argued for this in Kuczynski
(2005c). But we can leave that aside here.
15. Frege did not, at least not in this context, make any distinction between types and tokens.
He did not distinguish between tokens of (2) and (2) itself; and he did not distinguish, at least
not in this context, between tokens of “the richest man in America” and the type itself. Frege was
aware of the type-token distinction in general, and he had some keen insights into it. See, in
particular, his essay «The Thought» (Frege 1918). But that distinction has no significant role in
the theories of his that we are about to present.
16. See McCawley (1981: 22-56), especially pp. 22-23:
“[T]his chapter will be concerned with the “syntax” of quantifiers: elements of logical structure
that correspond to words like all, some, and most, which say which or how many of some set of
things have a given property.”
I don’t believe that McCawley’s definition is quite right. An utterance of “that man over
there is bald” says that a certain member of a certain set (the set of men in a certain area) has a
certain property (the property of being bald). So in virtue of having the form F…that man over
there…G that sentence-token says which member of a certain set has a certain property, and thus
qualifies as a quantifier according to McCawley’s definition. But, as Kaplan (1989) established,
“that man over there” is not a quantifier.
Nonetheless, McCawley is right to say that a quantifier is an expression that says how many
members of a certain set have a certain property.
 Conceptual Atomism and the Computational Theory of Mind

anything less than the idea that “the man over there” (or “the richest man in America”)
is a quantifier.17
Given Frege’s view as to what is meant by “the richest man in America is smart”,
along with his view that “the richest man in America” is a referring term, Frege must
deny the veritable truism that a quantifier is any expression such that, if a sentence S
contains that expression, S ipso facto answers (or comprises a sentence that answers)
some question of the form: How many phi’s are psi’s? (“No person is tall” answers the
question “how many people are tall?”, the same being true of “all people are tall” and
“one person is tall.” The sentence “if no person is tall, then no person is happy” does
not answer the question “how many people are tall?” or any other question of that
form. But it contains a sentence that answers that question, and is therefore no coun-
terexample to our definition.) It follows that Frege’s view is only as plausible as the
denial of this plausible, if not truistic, view as to what quantifiers are.
We have argued that “the richest man in America” is a Russellian quantifier if, as
Frege held, “the richest man in America is smart” is logically equivalent with “some-
thing x is uniquely a richest man in America; moreover, x is smart.” But one could re-
ject this argument on the following grounds. “Bill Gates is smart” is logically equiva-
lent with “something x is uniquely identical with Bill Gates; moreover, x is smart.” So
by reasoning parallel to ours, “Bill Gates” must be a quantifier. But, of course, “Bill
Gates” is not a quantifier. Our argument must therefore be rejected, since a parallel
argument establishes a falsehood.
How are we to deal with this problem? First of all, “Bill Gates is smart” is equiva-
lent with the just-mentioned existence-claim only if it is granted that Bill Gates is iden-
tical with some one individual. Though correct, this premise is extra-semantic in na-
ture. Somebody who said “Bill Gates is identical with several different people” would
not be guilty of linguistic incompetence, even though he would be guilty of failing to
register an obvious analytic truth. (Adherents of a certain widely held faith hold that
God is numerically identical with three different entities. Such people cannot categor-
ically be described as linguistically incompetent, even though they can categorically be
described as guilty of accepting an analytic falsehood. More generally, not all analytic
truths are strictly semantic. Given that x is knowledge, it follows analytically that x is a
case of justified belief. But it is not a matter of semantics alone that “if x is knowledge,
then x is a case of justified belief ” is true. In times past, some English-speaking phi-
losophers argued that, indeed, there are cases of knowledge that are not cases of justi-
fied belief. Those philosophers were not, on that account, guilty of not speaking Eng-
lish correctly: they knew what they wanted to say, and they were using the

17. An argument very similar to this is given by Russell (1912). Barwise and Perry (1983, 1999)
also present this argument, or one like it, in a clear and persuasive fashion; and, on the basis of
this argument, they have made a powerful case for some significant semantic positions. But the
philosophical community has been curiously unreceptive to their efforts. For example, Neale’s
(2005) book Facing the Facts only references it in passing, even though it is clear that, if cogent,
that argument is completely devastating to practically the entirety of Neale’s discussion.
 Chapter 4.  Some semantic consequences or our analysis 

English-language to say it. Their problem wasn’t that they were linguistically incompe-
tent; their problem was that their view was wrong. We will talk more about the distinc-
tion between semantic and analytic truth when, in Chapters 13 and 16, we discuss the
concept of formal truth.) Thus, even though “Bill Gates is smart” is logically equivalent
with “something x is uniquely identical with Bill Gates; moreover, x is smart”, that
equivalence is not strictly a matter of semantics.
But it is strictly a matter of semantics that “something x is uniquely identical with
Bill Gates; moreover, x is smart” is inconsistent with the proposition that Bill Gates is
identical with several different people. Thus, speaking from a strictly semantic view-
point, “Bill Gates is smart” is not logically equivalent with that existence-claim or, by
obvious extensions of these points, with any other quantified generalization, that being
why “Bill Gates” is not a quantifier. At the same time, supposing that Frege is right
about what is meant by “the richest man in America is smart”, it is a matter of seman-
tics alone that what is meant by that sentence is logically equivalent with the previ-
ously mentioned existence-claim.
The following argument makes it clear why this last point must be accepted. The
semantic rule for “the richest man in America is smart” is not adequately articulated
by the statement that:
(*) “the richest man in America is smart” encodes a proposition that is true iff the
richest man in America is smart.
After all, (*) is quite ambiguous, as it could mean either
(**) “the richest man in America is smart” encodes a proposition that is true iff the
person who in fact is the richest man in America is smart; in other words, for
some x such that x is uniquely a richest man in America, that sentence en-
codes a proposition identical or equivalent the singular proposition that x is
smart
or
(***) “the richest man in America is smart” encodes a proposition that is true iff,
supposing that somebody or other is a richest man in America is smart, any
such person is smart; in other words, that sentence encodes a proposition
identical or equivalent with the generalization that somebody is uniquely a
richest man in America and any such person is smart.
(**) is the view of somebody who believes that definite descriptions refer directly, and
is thus the antithesis of Frege’s view. (***) coincides with Frege’s view, being that of
somebody who believes that definite descriptions refer indirectly. But notice that (***)
is identical with the rule which, according to the Theory of Descriptions, gives the se-
mantics of “the richest man in America is smart.” Of course, the essence of the Theory
of Descriptions is that definite descriptions are quantifiers, not referring terms. This
 Conceptual Atomism and the Computational Theory of Mind

confirms our point that it is arbitrary and, in fact, incoherent to distinguish between
indirect reference, so-called, and quantification.
In any case, there is no strictly semantic equivalence between “Bill Gates is smart”
and the previously mentioned existence-claim; and it is therefore wrong to say that an
argument parallel to ours establishes the falsehood that “Bill Gates” is a quantifier.
It is not hard to find additional support for the view that Fregean “reference” is
really just quantification. In the aftermath of Kripke, we have good reason to believe
that, for some x (such that, as it happens, x is identical with Bill Gates), the literal
meaning of a token of (2) is simply:
(2LM) x is smart.
In the aftermath of Kaplan (1975, 1989), we have good reason to believe that an occur-
rence of:
(9) “that guy over there is smart”
encodes that same proposition provided that the demonstrative expression refers to
Bill Gates.
More generally, we have seen that a proper noun or demonstrative expression E refers
to O exactly if, in virtue of having the form FE has phiG, a sentence-token encodes a prop-
osition that is true exactly if O has phi – it being irrelevant what other properties O has.
So we have seen that, at least where proper nouns and demonstratives are concerned,
reference is mere labeling. It isn’t describing or conceptualizing: it is mere picking out.
In light of these points, suppose that Frege is right. Suppose, in other words, that
the occurrence in (2) of “the richest man in America” refers to Bill Gates, but that (2)
is logically equivalent with (2P). Notice that Bill Gates is not a constituent of (2P). What
is such a constituent is some concept that he uniquely instantiates. (To “instantiate” a
concept is to fall under it. I instantiate the concept of tallness because I am tall.) Fur-
ther, notice that Bill Gates is a constituent of the proposition literally meant by:
(1) “Bill Gates is smart.”
So if Frege is right about definite descriptions, then there are two fundamentally differ-
ent kinds of reference. There is the kind where, in virtue of containing a referring term,
a sentence has for its literal meaning a proposition that has the referent of that term as
a veritable constituent. (This would be Kripke-Kaplan “direct reference.”) And there
would also be a kind of reference that didn’t meet this condition. (This would be Frege-
style “indirect reference.”) If a sentence contained an expression E that “referred” in the
latter sense to O, then in virtue of that fact that sentence would encode a proposition
had as a constituent, not O itself, but some concept C such that O uniquely instantiates
C; and that proposition would be to the effect that (inter alia) C was uniquely instanti-
ated. Thus, if E “refers” to O, in the Fregean (indirect) sense of “reference”, there is some
concept C such that O uniquely instantiates C, and the proposition meant by FE has phiG
is to the effect that C is uniquely instantiated and, further, that any instance of C has phi.
 Chapter 4.  Some semantic consequences or our analysis 

So that proposition would semantically encode the very information provided by the
proposition that, according to Russell’s theory, is meant by Fthe C has phi. G But, in that
case, it is deeply unclear why E isn’t simply a Russell-style generalized quantifier.
For example, if Frege is right, then (2) semantically encodes exactly the same in-
formation as (2P), the reason being that “the richest man in America” refers indirectly
to Bill Gates. But there would be expressions that referred in a very different way to Bill
Gates. An example would be “Bill Gates” itself. That expression uncontroversially re-
fers to Bill Gates. But (1) is not logically equivalent with (2P), and the information se-
mantically encoded in (1) is not the same as that semantically encoded in any quanti-
fied generalization. (Of course, “Bill Gates is smart” is logically equivalent with
“something x is uniquely identical with Bill Gates; moreover, x is smart.” But we’ve al-
ready seen why neither this quantified generalization, nor any other is semantically
encoded in “Bill Gates is smart.”)
At the same time, if Frege is right about definite descriptions, then (2) has a great
deal in common with (2P). Both answer exactly the same question, namely: how many
richest men in America are there, and are they smart? Further, if Frege is right, the
semantic rule for (2) is: “the richest man America is smart” is true iff there is exactly one
richest man in America, and any such person is smart. That proposition is, of course,
identical with the semantic rule for (2P). It is once again unclear how Frege’s view dif-
fers from the view that “the richest man in America” is a quantifier.
Frege uses the word “reference” (“Bedeutung”) to describe the relationship be-
tween “the richest man in America” and Bill Gates. But, leaving aside that purely verbal
fact, Frege’s view is not meaningfully different from Russell’s.
In fact, Frege’s theory seems to be nothing more than a confused version of Rus-
sell’s theory.18 Russell (1917: 152–167) held that if E actually refers to O, then in virtue
of having the form F…E…G a sentence (or sentence-token) encodes a proposition that
has O itself as a constituent, as opposed to some sense or concept of O. Russell held
that if “Socrates” actually referred to Socrates – if it were not a quantifier that, as Rus-
sell puts it, merely “simulated” reference to Socrates – then for some x (such that x is
identical with Socrates) FSocrates has psiG would encode the singular proposition: x
has psi. Russell had the concept of what is now sometimes called “singular” or “direct”
reference. The relationship between an expression that refers in this sense to Bill Gates
is very different from the relationship that, according to Frege’s theory, holds between
“the richest man in America” and Bill Gates. The latter relationship is such that what is
meant by “the richest man in America is smart” is identical, or at least equivalent, with
a quantified generalization (namely, something is uniquely a richest man in America,
and any such thing is smart). The former relationship is such that, for some x (such that

18. McDowell (1998: Chapter 13) denies exactly this claim. But his justification for his rejection
of this is predicated on the correctness of his extremely strong form of content-externalism. We
have found reason to doubt whether any kind of content-externalism is correct, let alone Mc-
Dowell’s extreme version of it. We will continue to find reasons to question content-externalism.
 Conceptual Atomism and the Computational Theory of Mind

x is identical with Bill Gates), “Bill Gates* is smart” means: x is smart, where “Bill
Gates*” is a hypothetical term that refers, in the strict Russellian sense, to Bill Gates.
Because Russell (1912, 1917: 152–167) saw this, he rightly held that the second
kind of relationship has few or no meaningful similarities with the kind of relationship
that, if Frege’s theory is right, obtains between “the richest man in America” and Bill
Gates. Russell clearly saw that Fregean “reference” – what would now be called “indi-
rect” reference – is not significantly different from quantification.
Russell agreed with Frege on one important issue. He held that each of “Socrates is
smart” and “the richest man in America is smart” encodes a proposition identical, or
at least logically equivalent, with one of the form: something uniquely has phi, and any-
thing having phi is smart. But Russell held that, for this very reason, “Socrates” and “the
richest man in America” are quantifiers, as opposed to devices of reference; for he saw
that describing them as terms that “referred” to Socrates and Bill Gates would obscure
the important distinction to be made between instances of quantification and instanc-
es of what are clearly reference. Russell’s theory of descriptions is thus a clear and co-
herent version of Frege’s theory. Unlike Frege, Russell does not describe as “reference”
what is in fact quantification. He knew that the relationship that “Bill Gates” bears to
Bill Gates has no meaningful similarities to the relationship that, if Frege’s theory is
correct, either “the richest man in America” or “Bill Gates” bears with respect to Bill
Gates. And he knew that the latter relationship has many meaningful similarities with
the relationship that, if his own Theory of Descriptions is correct, either “the richest
man in America” or “Bill Gates” bear to Bill Gates. He thus saw that referring to the
second relationship as one of “reference” would merely obscure a crucial distinction
(that between singular reference and quantification), and would also hide a profound
kinship (that between Fregean “reference” and quantification). So Russell used the
term “quantification” where Frege had used the term “reference.” But apart from that
(and apart from the incoherences in Frege’s semantics that we are about to discuss), the
semantics that Russell provides for proper nouns and definite descriptions is the same
as that provided by Frege. In any case, it is hard to find any real difference.
Russell’s conception of reference was entirely correct. His one fault was that he
misapplied that conception. For he thought that actual cases of reference were almost
never found in everyday discourse. (In his view, one could refer to a content of one’s
consciousness – by using a token of “this” or, conceivably, a proper noun of one’s invent-
ing. But he thought that nothing apart from a content of one’s own consciousness could
be the object of reference. In Chapter 9, we will discuss at great length how this view
involves a misunderstanding of the concept of language. But it is noteworthy that Rus-
sell held that, if there are any genuinely referring expressions, certain demonstrative-
expressions are among them. This is a striking anticipation of Kaplan’s (1989) theory.)
Accordingly, Russell denied that “Socrates” refers to Socrates, and he thought that
the former was a quantifier. He thought that, in virtue of containing an occurrence of
“Socrates”, a sentence-token encoded a proposition that had for a constituent, not Soc-
rates himself, but rather some concept that Socrates instantiates. He held the same view
 Chapter 4.  Some semantic consequences or our analysis 

with regard to “the great philosopher who drank hemlock.” These views, generalized to
all proper nouns and definite descriptions, constitute Russell’s Theory of Descriptions.
We’ve seen why that theory is false, or at least unnecessary, even though the con-
ception of reference that underlies it is unexceptionable. We’ve also seen why Frege’s
theory is not appreciably different from it. It follows that Frege’s theory is wrong for the
same reasons as Russell’s. But we will now see that, because Frege conflated reference
with quantification, his theory, unlike Russell’s, is incoherent, and not just false.
If Frege is right, i.e. if (2) and (2P) are logically equivalent, then (2) will be true in
exactly the same circumstances as (2P). Of course, (2P) has a truth-value regardless of
whether there is a unique richest man in America. If there is no such man, then (2P) is
false, since it affirms that there is such a man. But according to Frege, (2) is without
truth-value if there is no richest man in America. This is flatly inconsistent with the
idea that (2) is logically equivalent with (2P); and it provides evidence in favor of our
analysis. So far as you are affirming that such and such exists, you are not referring to
it. Surely I am not referring to the number two in saying “there is a unique even prime
x and x is even.” (In any case, I am referring to it only in the loose, vernacular sense of
the word “refer.” I am not referring to in the strict, philosophical sense of the term – the
sense in which I would be referring to it if I were to say “two is an even number.”)
Where there is an affirmation of existence, there is no reference, as Strawson (1950)
made clear. If Frege is right to say that (2) and (2P) are logically equivalent, then an ut-
terance of (2) is simply false – not lacking in truth-value – if there is no unique richest
American man; and the reason for this would be that, if there is no such man, then
what (2) has affirmed is false. Given that affirming a thing’s existence is different from
referring to it, it follows that “the richest man in America” is not a singular term if, as
Frege holds, (2) and (2P) are logically equivalent.
Frege seemed to be aware, on some level, that referring and affirming existence are
different things. That, presumably, is why he held that (2) is without truth-value if
there is no unique richest man in America. But if (2) is without truth-value under that
circumstance, then Frege’s other contention – that (2) and (2P) are logically equivalent
– is simply false. Frege’s views on reference are thus incoherent.
Let us momentarily backtrack; let us consider some objections to the argument
given a few pages back (the argument to the effect that Frege conflated reference with
quantification):
You say that, if Frege is right about (2), then there are two totally different kinds of
reference. But didn’t we already know that? There is “rigid” and “non-rigid” desig-
nation, to use Kripke’s (1972) terms. “Bill Gates” rigidly refers to Bill Gates, and
“the richest man in America” non-rigidly refers to him. Obviously rigid designa-
tion is the same thing as direct reference, and non-rigid designation is the same
thing as indirect reference. So where is the problem with Frege’s theory?

Direct reference is not the same thing as rigid designation. Consider a demonstrative
expression, e.g. “that man over there.” It refers to different people on different occasions.
 Conceptual Atomism and the Computational Theory of Mind

A fortiori it refers to different people in different worlds. It is thus a text-book case of an

expression whose referent varies depending on the circumstances, and would thus seem
to be a paradigm of non-rigid designation. But it is directly referential.
In response, one might argue that “that man over there” is a rigid designator, on
the grounds that it refers directly. But this would either be a purely terminological
move – and a confusing one at that – or a case of begging the question. The referent of
“Bill Gates” is not context-sensitive. Given the rules of English semantics, its reference
is fixed. “Bill Gates” is also a paradigm-case of a rigid designator.19 The referent of “that
man over there” is context-sensitive. Given only the rules of English semantics, its
reference is not fixed.20 If we say that “that man over there” is a rigid-designator, then
we are saying that rigidity and non-rigidity are not to be understood in terms of easily
ascertained and uncontroversial facts about context-sensitivity, and that they are to be
understood in terms of extremely controversial and hard to ascertain facts about the
constituencies of propositions. If we say that “that man over there” is rigid, on the
grounds that it is directly referential, then the terms “rigid” and “non-rigid” cannot be
used without prejudging the outcomes of highly theoretical semantic and metaphysi-
cal debates, and they thus cease to have any descriptive utility. So it is not a reasonable
option to define “rigid” as “directly referential.” “Rigid” must continue to be defined as
“such that what it refers to does not depend on the context of utterance.”
But even if we set these points aside, the objector’s point fails. Suppose that “the
richest man in America” is indeed a “non-rigid designator” and that Frege is right to
think that (2) and (2P) are logically equivalent. In that case, given the arguments al-
ready presented, it isn’t clear why “non-rigid designation” isn’t just a form of quantifi-
cation. In any case, as we’ve seen, “non-rigid designation” would have very little in
common with any clear-cut case of reference, and it would have much in common
with quantification.
As we saw, if Frege is right to hold that (2) and (2P) are logically equivalent, then
the relationship between “Bill Gates” and Bill Gates is very different from the relation-
ship between “the richest man in America” and Bill Gates. It is not hard to extend the
argument that we gave in defense of this. The proposition meant by (1) is true only in
worlds where Bill Gates exists. But the proposition meant by (2P) is true in some worlds
where he does not. The former is true in a world iff Bill Gates is smart in W. The latter
can be true in a world W where Bill Gates is not smart. The relationship between “Bill
Gates” and Bill Gates is indisputably a case of bona fide reference. But – supposing,
with Frege, that (2) and (2P) are logically equivalent – that relationship bears little re-
semblance to the relationship between “the richest man in America” and Bill Gates.
Of course, we might use the expression “designation” (or “reference”) to refer to
both relations (with the qualification that reference was “rigid” in the one case and
“non-rigid” in the other). But that would only mean that our terminology masked

19. Kripke (1972: Lecture II).

20. Kripke (1972: Lecture II).
 Chapter 4.  Some semantic consequences or our analysis 

crucial semantic and logical differences. It would mean that “non-rigid reference”
would denote a relationship that, as we previously saw, was virtually indistinguishable
from quantification.
Let us now state our second criticism of Frege. It will help if we begin by consider-
ing two other paradigms of ours:
(s2) “Smith thinks that the richest man in America is smart.”
(s7) “Smith thinks that the first person to make a fortune by selling personal com-
puters is smart.”
As we’ve discussed, Frege took it for granted that definite descriptions are singular
terms. Frege also took it for granted that (s2) and (s7) have different literal meanings
and, further, that they have (or, depending on people’s beliefs, could have) different
truth-values. Finally, and most importantly, Frege took it for granted that the under-
lined parts of (s2) and (s7) refer to different propositions. (Given his other views, it was
mandatory that he hold this.)
Here is where the problems begin. Suppose that we accept the principle of compo-
sitionality. In that case, the underlined part of (s2) must co-refer with the underlined
part of (s7), supposing that the definite descriptions in them co-refer. So the underlined
part of (s2) must refer to the same proposition as the underlined part of (s7). But accord-
ing to Frege, the underlined parts refer to different propositions, that being why (s2) and
(s7) may differ in truth-value. Frege deals with this by saying that, in (s2) and (s7), the
definite descriptions do not refer to Bill Gates (or any other individual), but rather refer
to their “senses.” (Given his acceptance of compositionality, along with his other se-
mantic views, Frege had no choice but to take this position.) For exactly similar reasons,
the underlined expression in:
(10) “Jones thinks that Smith thinks that the first person to make a fortune by sell-
ing personal computers is smart,”
refers not to Bill Gates, but to the sense of the customary sense of “the first person to
make a fortune by selling personal computers.”
Frege’s view, appropriately generalized, is this. If a definite description (or proper
noun) occurs in the context of an expression denoting a proposition, that definite de-
scription refers to its “sense.” So in
(11) “it is true that the first person to make a fortune by selling personal computers
is smart,”
the occurrence of the definite description does not refer to Bill Gates. Rather, it refers to
what would ordinarily be the “sense” of that definite description (the concept first person
to make a fortune by selling personal computers). Given obvious extensions of this reason-
ing, Frege’s view demands that the occurrence of the definite description in:
 Conceptual Atomism and the Computational Theory of Mind

(12) “it is true that it is true that the first person to make a fortune by selling per-
sonal computers is smart,”
refers to the sense of the sense that it ordinarily has.
But, as Davidson (1980b: 93–108) points out, this view is deeply implausible. And
is obviously ad hoc to suppose that the occurrence of “the first person to make a for-
tune by selling personal computers” refers to one thing in (12), a different thing in (11),
and a third thing in (1).21
There is another problem. What is the sense of the sense of the sense of “the first
person to make a fortune by selling personal computers”? The sense associated with
“the first person to make a fortune by selling personal computers” is not hard to iden-
tify: it is the concept first person to make a fortune by selling personal computers. But the
sense of that sense (not to mention the sense of that sense of that sense) is not so easy
to identify. (Here we must distinguish between senses and modes of presentation. A
mode of presentation is how an object is given to somebody who is perceiving or
thinking about that object. More precisely, a mode of presentation of x is some concept
or body of information C such that x satisfies C and such that, consequently, if one
grasps C, one thereby grasps x. Modes of presentation are not linguistic entities. After
all, when a non-linguistic creature sees the moon, the latter is given to the former
through a mode of presentation. Senses, on the other hand, are linguistic entities. More
precisely, the sense of an expression R that refers to x is a mode of presentation of x that
is meant by R. But, in this context, I will use the term “sense” to cover both senses and
modes of presentation, since the consequent loss of precision is more than offset by the
gain in expository simplicity.)
In fact, it seems unlikely that there is any such thing as the sense of that sense (or,
therefore, of the sense of that sense). Any given spatiotemporal entity can be given
through infinitely many different senses. If the same is true of concepts, then there is no
such thing as the sense of the sense of “the first person to make a fortune by selling
personal computers.” Frege would then have to pick some specific sense out of these
infinitely many. But any such selection would be completely arbitrary and couldn’t pos-
sibly correspond to any fact about English-semantics.
To block this, Frege would have to say either (a) that he was not concerned with
the actual semantics of English or (b) that, given any concept, there is some one sense
through which it must be grasped. Let us consider each option.
(b) is false. A given concept can be grasped in different ways, i.e. through different
“modes of presentation.” In fact, this is a point that Frege himself emphasizes in the
Foundations of Arithmetic (Frege 1884), and we may borrow one of Frege’s own exam-
ples to illustrate it. Consider the concept of the number zero. There are different ways

21. Davidson (1980b: Chapter 1) has ably argued that Frege’s position is incompatible with the
fact that English (and other natural languages) learnable. Given how effectively Davidson presents
his argument, I will not repeat it, and will merely refer to the reader to Davidson’s article. In Kuc-
zynski (2005c), I deal with some of the objections that might be made to Davidson’s argument.
 Chapter 4.  Some semantic consequences or our analysis 

of thinking about that concept. It can be thought of as the concept that is instantiated
by the predecessor of one or as the concept that is instantiated by the successor of
negative-one, and so on. Also, given how hard it is to produce a correct analysis of that
concept, it would be artificial to suppose that there were only one way of thinking
about it, the reason being that, were that the case, grasping that concept would be an
all or nothing affair: one would either be oblivious to it or one would have complete
insight into its structure. But, as Frege (1884) himself emphasized, one can typically
learn more about a concept that one already grasps, just as one can learn more about
the structure of a physical object that one sees. This suggests that concepts, like spatio-
temporal objects, may be given through indefinitely many different modes of presenta-
tion. If correct, this means that (b) is false.
If option (a) is chosen, then Frege’s analysis is ex hypothesi irrelevant to what is
meant by any actual (English or German or Japanese…) expressions, and is thus ir-
relevant to our hypotheses as to what such expressions mean. This is not to mention
that Frege does seem to be saying quite explicitly that in the English sentence “Socrates
is tall and Smith believes that Socrates is tall”, the two occurrences of “Socrates” do not
co-refer, as one of them refers to a person and the other to a mode of presentation.
Russell’s theory does not have any of these problems. For any number n, if an occur-
rence of Fthe phiG is preceded by n operators, there is a well-defined rule saying how it
parses out.22 That rule is: F…the phi…G means that (or is true iff): exactly one thing O
has phi, and…O…Equivalently, that rule is: Fthe phi has psiG means that (or is true iff):
exactly one thing O has phi and, moreover, O has psi. As we saw, if Russell is right, there
will be cases where an occurrence of Fthe phiG parses out in different ways. (This will be
the case whenever it falls within the scope of any operator that forms molecular propo-
sitions out of atomic ones, as Kaplan (1975b) points out.) So, to use our old example, if
Russell’s theory is right, there are different ways that the definite description in (s2) can
parse out. But this is not to the discredit of Russell’s theory, since the propositions that
are in fact conveyed by s2 correspond to those different ways of eliminating the definite
description.
But even though Russell’s theory doesn’t have the problems that we’ve just attrib-
uted to Frege’s, it does have other, non-trivial problems, as we’ve seen. At the same
time, our theory does not have those problems, and it also appears to explain the rel-
evant facts about cognitive significance.

22. In any case, this is the traditional view. See Neale (1990).
But, in actuality, what was just said is true only of indicative sentences containing definite
descriptions. Given a sentence or imperative containing a definite description, Russell’s theory has
the consequence that it is, on at least one disambiguation, ill-formed and thus without any cohe-
rent meaning. Neale (1990) considers only indicative sentences, and thus doesn’t confront this
problem. In Kuczynski (2004d), I show that Russell’s theory wrongly predicts that any non-indica-
tive sentence containing a definite description is ill-formed on at least one disambiguation.
 chapter 5

Modality, intensionality, and

a posteriori necessity

In the previous chapter, we saw how a failure to distinguish token-meaning (seman-

tics) from type-meaning (pre-semantics) has led to some highly implausible semantic
views. We also saw how reasonable alternatives to those views are generated if that
distinction is taken into account. In this chapter, we will see how some erroneous
theories about modality and conception result from a failure to distinguish semantics
from what might be called “pre-pre-semantics.”
Throughout this chapter we must keep in mind our point that it is always sentence-
tokens that bear propositions, and that sentence-types never do so. The semantic con-
tent of sentence-type “Smith is tall” is a rule that assigns a proposition to any given to-
ken of that type. That is why different tokens of that type can have different truth-values.
(Such a token is false if uttered during Smith’s infancy and true if uttered during his
adulthood.) Of course, any two tokens of “one and one make two” have the same truth-
value. But that is to be explained in terms of the non-semantic fact that mathematical
reality doesn’t change; it is not to be explained by supposing that the present-tense
marker is semantically ambiguous between temporal and non-temporal meanings.1
As we’ve discussed, Kripke (1972) made it clear that “Hesperus” and “Phospho-
rous” are not quantifiers and that they are not disguised descriptions. Kripke made a
related point that we have not yet explicitly discussed, namely: proper names (e.g.
“Hesperus”, “Socrates”) are rigid designators. In other words, given the semantics of
English, it is fixed what those expressions (or, strictly speaking, their tokens) refer to.
Let us develop this point. (To expedite discussion, I will speak as though expression-
types refer. The inaccuracies created by this way of speaking are easily eliminated.)
Imagine a world W where Al Gore, as opposed to George W. Bush, is U.S. Presi-
dent in 2006. Given only that (in 2006) people in W refer to Gore as “the U.S. presi-
dent”, it does not follow that there is a language in W whose semantic rules differ at all
from those of English. But suppose that, in at least one of the languages in W, “Hespe-
rus” refers to Mars, as opposed to Venus. In that case, we can conclude that there is
some language in W that does not have precisely the same semantic rules as English.
A language that is just like English except that, in it, “Plato” refers to Socrates and
“Socrates” refers to Plato, does not have precisely the same semantic rules as English.

1. See Kripke (1977) for a classic discussion of why facts about cognitive significance shouldn’t
be explained by positing semantic ambiguities if there is an available non-semantic explanation
of those facts.
 Conceptual Atomism and the Computational Theory of Mind

Strictly speaking, that language isn’t English.2 For similar reasons, a language in which
“Hesperus” refers to Mars, not Venus, is not English, even though it may otherwise be
just like English.
Of course, “Hesperus” and “Phosphorous” both (rigidly) refer to Venus. It follows
that neither “Hesperus” nor “Phosphorous” can fail to refer to Venus in any language
that has the same semantic rules as English.
Here is where matters become interesting. Obviously
(1) “Hesperus is Hesperus”
expresses a trivial proposition. But
(2) “Hesperus is Phosphorous”
expresses a non-trivial proposition. Frege and Russell dealt with this by saying that
“Hesperus” and “Phosphorous” are quantifiers or disguised descriptions. “Hesperus” is
synonymous with “the first celestial body to appear in the evening sky” and “Phospho-
rous” is synonymous with “the last celestial body to disappear from the morning sky.”
So according to Russell, (2) means (or is logically equivalent with):
(2R) Exactly one thing x is a first celestial body to appear in the evening sky; and
exactly one thing y is a last celestial body to disappear from the morning sky;
and nothing that has the property of being a first celestial body to appear in
the evening sky lacks the property of being a last celestial body to disappear
from the morning sky.
Frege’s position is not different from Russell’s, at least not in any way that is relevant in
this context.3
We know from Kripke that Frege and Russell were wrong. “Hesperus” is not a de-
scription. And “Hesperus”, unlike “the first celestial body to appear in the evening sky”,
rigidly refers to Venus. Of course, the same is true of “Phosphorous.” Each of the refer-
ring terms in (2) is a label; and the structure of the proposition encoded in (2) is not a
quantified, Russell-style generalization. Rather, that proposition has the form x=y.

2. In any case, there is one important disambiguation of the term “English” relative to which
that language isn’t English. From a syntactician’s or phonetician’s standpoint, the two languages
are not significantly different. But here we are doing semantics, not syntax or phonetics. In
Chapter 16, we will discuss the various possible delineations of the term “English” and, thus, of
“Chinese”, “French”, and so on.
3. McDowell (1998) says that Kripke’s arguments falsely presuppose that Russell’s and Frege’s
views are ultimately the same. McDowell’s argument for this is tethered to his content-externa-
lism, which we have seen to be false. In Chapter 7, we will discuss the underpinnings of Mc-
Dowell’s claim at length.
 Chapter 5.  Modality, intensionality, and a posteriori necessity 

There is one last point to make before we can close Kripke’s argument. The relation
of identity holds necessarily.4 Nothing could possibly fail to be itself. Nothing could
have been a numerically different object. This point can be established formally. Sup-
pose that x is only contingently identical with y. Obviously x isn’t contingently identi-
cal with x. Thus, if x is only contingently identical with y, then y has some property that
x does not have (y has, and x lacks, the property of being contingently identical with
x). So by Leibniz’s Law, x and y must be distinct. When it holds, identity holds neces-
sarily. Any true statement of the form x=y is necessary.
Let us now close Kripke’s argument. (2) expresses a true proposition of the form
x=y. That proposition holds necessarily, as we just saw. But (2) is obviously an a poste-
riori statement. It expresses a synthetic, not an analytic, claim. Empirical work was
needed to establish the truth of (2). There are thus necessary a posteriori truths. Not all
necessity is analytic (or a priori).

A posteriori necessary truth: a reassessment

The argument just presented involves a major non-sequitur and is incorrigibly fallacious.
We’ve seen good reason to believe that, for some x, an utterance of “Smith is tall” has for
its literal meaning the singular proposition x is tall. We’ve also seen reason to believe that
what is communicated by any such utterance is much richer than that singular proposi-
tion. What is communicated by such a proposition is a function not only of its literal
meaning, but also of the information through which one ascertains its literal meaning.
That information is existential, as we have seen. You cannot just see Smith. You must see
a thing that is embedded in a situation any perceptual representation of which is replete
with descriptive content. And when someone speaking to you points to someone and
says “that guy is Smith”, the information that enables you to associate “Smith” with the
right object has the form: there is some entity x such that x has such and such properties
(e.g. x has the property of now being pointed out by the person with whom I am now talk-
ing) and such that “Smith” names x. Because “Smith” has narrow-scope with respect to
the descriptive information (there is an entity having such and such properties), it is a
mere label, and not (pace Russell and Frege) a description or quantifier. But, as we have
stressed, even though it has been semantically neutralized, by virtue of being given wide-
scope in the relevant existence-claim, that descriptive-existential information still re-
mains operative: you access the meaning of “Smith” through that existence-claim (or, at
any rate, through some other existence-claim that “interlocks” with it in the way de-
scribed in Chapter 3). For this reason, what an utterance of “Smith is tall” communicates
to you will be an existence-claim – it will have the form somebody x uniquely has such and
such properties; moreover, “Smith” names x, and x is tall. (Of course, it is not always

4. The argument about to be given is found in Kripke (1972). But it is actually due to Quine
(1953: 155).
 Conceptual Atomism and the Computational Theory of Mind

through ostensive definition that one learns what proper names refer to. But given what
we saw in Chapter 3, it is clear how the points just made can be extended to deal with
cases where word-reference is learned non-ostensively.)
Of course, what we just said about “Smith” is true of “Hesperus” and “Phosphorous.”
For illustrative reasons, let us suppose that you learned the meanings of those terms os-
tensively. Your friend Jones point to a certain object in the morning sky, and says “that is
Hesperus.” The object indicated is quite clearly the last celestial body to disappear from
the morning sky. (Of course, given only that some object is the last celestial body to dis-
appear from the morning sky on one occasion, it doesn’t follow that it always has that
property, i.e. it doesn’t follow that tomorrow some other celestial body won’t have that
property. But let us suppose that, on the basis of the experience just described plus your
knowledge of astrodynamics, you correctly infer that the object indicated must be the
last celestial body to disappear from every morning sky.) There is some x such that the
semantic rule for “Hesperus” is simply:
(HS) “Hesperus” refers to x.
(More accurately, that rule is: tokens of “Hesperus” refer to x, for the appropriate value
of x. This nicety is important, as we will see.) But (HS) is given to you through some
claim having (approximately) the form:
(HE) There is some object x such that x is a unique last celestial object to disappear
from the morning sky, and “Hesperus” names x.
A consequence is that if somebody says to you:
(HL) “Hesperus is lovely”,
what is communicated to you is not (merely) x is lovely, for some x, but is rather:
(HLG) There is some object x such that x is a unique last celestial object to disappear
from the morning sky; moreover, “Hesperus” names x, and x is lovely.
For reasons that we discussed in Chapter 3, this doesn’t mean that you wrongly take
tokens of (HL) to have (HLG) for their literal meanings. You know that the literal
meaning of an utterance of (HL) is confined to the underlined part of (HLG). That is
why you have the right modal and semantic intuitions; that is why you know that what
is literally meant by “Hesperus is the last celestial body to disappear from the morning
sky” is not analytic; and that is why you know that “Hesperus might not have been the
last celestial body to disappear from the morning sky” is not absurd and does not de-
scribe an impossible situation. But it remains a fact that you access the meaning of to-
kens of (HL) through (HLG) and that such tokens thus communicate (HLG) to you.
Given any predicate phi, exactly similar points hold with regard to what is communi-
cated to you by FHesperus has phi.G
Once again, you and Jones are gazing at the sky. He points to a celestial object and
says “that is Phosphorous.” The object indicated is quite clearly the first celestial object
 Chapter 5.  Modality, intensionality, and a posteriori necessity 

to appear in the evening sky. For exact analogues of the reasons just given, what is
communicated to you by a token of:
(PL) “Phosphorous is lovely”
will have approximately the form:
(PLG) There is some object x such that x is a unique first celestial object to appear in
the evening sky; moreover, “Phosphorous” names x, and x is lovely,
even though (as you know) what is literally meant by such an utterance is confined to
the underlined portion of (PLG). So while it is true that the semantics of “Phospho-
rous” is given by the rule:
(PS) “Phosphorous” refers to x,
the epistemology of that term is given to you through some piece of information like:
(PE) There is some object x such that x is a unique first celestial object to appear in
the evening sky, and “Phosphorous” names x.
As we’ve discussed, when we want to know what is communicated by statements, we
must look not at the propositions that give their semantics, but at the propositions that
give the epistemology of their semantics.
Given what we’ve said, it immediately follows that, for any phi, what a token of
F
Hesperus has phiG conveys to you is:
(I) There is some object x such that x is a unique last celestial object to disappear
from the morning sky; moreover, “Hesperus” names x, and x has phi.
It also immediately follows that what a token of FPhosphorous has phiG conveys to you
is:
(II) There is some object x such that x is a unique first celestial object to appear in
the evening sky; moreover, “Phosphorous” names x, and x has phi.
Given (I) and (II), it follows straightforwardly that what a token of
(2) “Hesperus is Phosphorous”
conveys to you will be:
(III) There is some object x such that x is a unique last celestial object to disappear
from the morning sky; moreover, “Hesperus” names x; there is some object y
such that y is a unique first celestial object to appear in the evening sky; more-
over, “Phosphorous” names y; moreover; x=y.
So what is conveyed by (2) will be exactly what Russell thought (wrongly) to be its lit-
eral meaning. What is conveyed by (2) is not a singular proposition of the form a=b. (I
 Conceptual Atomism and the Computational Theory of Mind

am switching variables for obvious reasons.) What is conveyed is a descriptively rich

existence-claim.
But, for reasons that we’ve discussed, what is literally meant is confined to the un-
derlined part of (III). And you know this. You know it for exactly the reasons that you
know that the literal meaning of a token of “Hesperus is lovely” is confined to the un-
derlined part of (HLG) (or, more accurately, that it is what is confined to a certain
specialization of the propositional-form corresponding to the underlined part.)
Of course, that underlined part is necessarily true. As we discussed, any genuine
identity claim holds necessarily, if it holds at all. But we don’t have any a posteriori
necessary truth here. Astronomical work was needed to ascertain the truth expressed
by (III). That proposition is entirely contingent.
Tokens of (2) do indeed encode necessary truths. For what is literally meant by
such tokens is a true proposition of the form a=b. But what is a posteriori is not what
is literally meant by (2) or tokens thereof, but is rather what is communicated by such
tokens. What is a posteriori is the information through which you access the literal
meaning of tokens of (2). And that information is contingent.
The literal meaning of (2), on the other hand, is analytic. That literal meaning is
given by a singular proposition of the form x=y. Because we are dealing with a singular
proposition, we are dealing with a structure that comprises the object x and the object y.
We are not dealing with a structure that comprises senses or descriptions of x or of y.
Apart from the concept of equality, only x and y – the objects, not any concepts thereof
– make it into that proposition. Since x is identical with y, some one object occupies the
place that (in our notation) corresponds to the occurrence of the “x” and the occurrence
of the “y.” And because, for the reasons just given, that object alone – without the accom-
paniment of any description or concept – is what occupies that place, the proposition in
question has the form x=x. And any proposition of that form is analytic.
Of course, empirical work is needed to grasp x, at least in this case. But for a prop-
osition to be analytic is not for it to be such that no empirical work is needed to grasp
it (or its constituents); it is for that proposition to be such that, once grasped, no em-
pirical work is needed to determine its truth-value.
So when we distinguish between semantics, on the one hand, and the epistemology
of semantics, on the other, we don’t have anything that is both necessary and a poste-
riori. We have a strictly analytic proposition (corresponding to the underlined part of
(III)) that is necessary; and we also have a contingent, a posteriori proposition (corre-
sponding to (III) in its entirety). The idea that we have unearthed anything that is both
necessary and a posteriori (or “synthetic”) embodies a failure to distinguish semantics
from pre-semantics – a failure to distinguish literal meaning from the information
through which we access it.
The concept of necessary, a posteriori truth is a conflation of two concepts: the
concept of literally meant, analytic truth and the concept of pre-semantically commu-
nicated (but not literally meant) contingent truth.
 Chapter 5.  Modality, intensionality, and a posteriori necessity 

Kripke provided an argument similar to the one just considered that purports to show
that what is meant by “water is H2O” and “light is electromagnetic radiation” express neces-
sary, a posteriori truths. But what we just said about (2) applies to these other cases.
This is not to say that there are no instances of necessary a posteriori truth. But given
only what Kripke says (in connection with “Hesperus is Phosphorous”, and other so-called
necessary a posteriori identities), it seems probable that the concept of necessary a posteriori
truth is yet another expression of the failure to distinguish semantics from pre-semantics.

Pre-semantics versus pre-pre-semantics

In the previous chapter, we saw how, where definite descriptions and demonstratives
are concerned, a failure to distinguish semantics from pre-semantics has led to some
false and, in fact, incoherent doctrines. In this chapter, we have seen yet another exam-
ple of that phenomenon. But here it becomes important to distinguish two very differ-
ent kinds of pre-semantics.
The semantics of the expression-type “that bald man over there is tall” is given by
a rule that assigns the singular proposition O is tall to a token t of that expression,
provided that t occurs in a context where some person O is uniquely a contextually
salient bald man. Because one must work through type-meaning to compute token-
meaning, what is communicated is not (merely) O is tall but is rather: somebody x is
uniquely a contextually salient bald man; moreover, x is tall. Because of this, many se-
mantic theories, e.g. those of Frege and Russell, involve a failure to distinguish type-
semantics from token-semantics.
But the semantic rule for “Hesperus” has nothing to do with the kind of descrip-
tive information embodied in existence-claims like:
(HE) There is some object x such that x is uniquely a last celestial object to disap-
pear from the morning sky, and “Hesperus” names x.
(HE) gives the information through which you grasp the semantic rule for “Hesperus.”
As a whole, (HE) does not give the semantic rule for the type “Hesperus” or for its to-
kens. Only the underlined portion of (HE) does that. There is some x such that x is
identical with Venus, and such that the semantic rule for “Hesperus” is simply:
(HS) “Hesperus” refers to x.
More exactly, there is some x such that the semantic rule for “Hesperus” is simply:
(HST) Tokens of “Hesperus” refer to x.
We’ve already discussed why the expression-type “Hesperus” doesn’t refer to anything, the
reason being that, for some x (such that x is identical with Venus), the semantic content
of the type “Hesperus” is given by a rule (a constant function) that assigns x to its tokens.
 Conceptual Atomism and the Computational Theory of Mind

Through the information borne by (HE), you learn the semantic content of the type “Hes-
perus.” That content is given by (HST). On the basis of your knowledge of (HST), you
learn what is meant by specific tokens of “Hesperus” and, therefore, what is meant by
particular tokens of “Hesperus is lovely”, “Hesperus is a planet, not a star”, and the like.
So as a whole, (HE) doesn’t state the semantics of the expression-type “Hesperus”:
only the underlined portion does that. Rather, (HE) states the information through
which you learn the semantic content of the expression-type “Hesperus.” That content,
in its turn, assigns a semantic content to tokens of that expression. So if token-meaning
is semantics, and type-meaning is pre-semantics, then what you learn through (HE) is
pre-pre-semantics.
Frege and Russell both thought that “Hesperus” had for at least part of its meaning
some concept like last celestial object to disappear from the morning sky. This view in-
volves several confusions. First, there is no such thing as “Hesperus” simpliciter. There
is “Hesperus”, the expression-type, and there are tokens of that type. For some x, the
semantic content of that expression-type is given by the proposition: tokens of “Hespe-
rus” refer to x. One must learn the semantic content of “Hesperus”, the expression-type,
through some highly descriptive claim – some claim like (HE). As a result, what is com-
municated by a token of “Hesperus is lovely” is some descriptively rich existence-claim.
So any token of “Hesperus is lovely” will, in virtue of containing an occurrence of “Hes-
perus”, communicate a proposition involving some concept like last celestial object to
disappear from the morning sky. But that concept has nothing to do with the literal
meaning of that sentence-token. Nor does it have anything to do with the literal mean-
ing of that sentence-type. So nothing having to do with that concept has to do with the
semantics or even the pre-semantics of that token (at least not if, by “pre-semantics”, we
mean what is meant by the corresponding type). Rather, that concept is part of the pre-
pre-semantics of that token. The concept last celestial object to disappear from the morn-
ing sky is, at most, part of the pre-pre-semantics of “Hesperus”; and propositions identi-
cal or equivalent with something x last celestial object to disappear from the morning sky,
and x is lovely are, at most, part of the pre-pre-semantics of tokens of “Hesperus is
lovely.” They aren’t part of the semantics of such tokens; and they aren’t part of the pre-
semantics of that token (i.e. they aren’t part of what is literally meant by the correspond-
ing type). Rather, they are part of the information through which one accesses the
meaning of the relevant-type; they are part of the information through which one ac-
cesses type-meaning (pre-semantics) and are thus pre-pre-semantic.

Frege’s theory a conflation of three different concepts

Here we need to qualify a point a just made. As we saw earlier, one often forgets the
information through which one initially learned the semantics of an expression. Sup-
pose that I learn who “Smith” refers to under the following circumstance. At a certain
time, in a certain place, I see a person with such and such properties; my companion
 Chapter 5.  Modality, intensionality, and a posteriori necessity 

points to Smith and says “that guy is Smith.” For reasons already discussed, what is
relayed to me is not some singular proposition of the form “Smith” names x, but is
rather one of the form: at such and such time and place, somebody x has such and such
properties and “Smith” refer to x (or, more accurately: tokens of “Smith” refer to x). Of
course, I may forget that particular existence-claim and still be able to think about and
refer to Smith. But so far as this possible, that is because my knowledge of that particu-
lar existence-claim has been replaced with a different one that interlocks with the first
(or with a third one that interlocks with the second…). As we saw in Chapters 2 and 3,
I must work through that existence-claim – that second or third or fourth…claim – to
compute the meaning of “Smith is a professor” (or, by exactly similar reasoning, to
compute the meaning of “Hesperus is a planet”). That existence-claim is, of course, no
part of the literal meaning of a token of “Smith is a professor.” Nor, obviously, is it any
part of the semantic content of the corresponding-type. That existence-claim consti-
tutes the information through which I access the semantics of the corresponding type.
(For some x, that semantics is given by the rule: tokens of “Smith is a professor” are
true iff x is a professor.) And on the basis of my knowledge of that rule, I compute the
meaning of a given token of that sentence-type. So the concept person having such and
such properties is no part of the semantics of any token of “Smith” or of the semantics
of the corresponding type (so it is no part of the pre-semantics, in the strict sense, of
such a token). Rather, that concept is, at most, part of the pre-pre-semantics of a token
of “Smith” and it is, at most, part of the pre-pre-semantics of a token of “Smith is tall”,
“Smith is a professor”, and so on. (I say “at most” because the concept person with such
and such properties may not be a part of the pre-pre-semantics of “Smith”, depending
on how one learns the semantics of “Smith.” If, while talking to me, somebody points
to a person with thus and such properties and says “that is Smith”, then the concept
person with thus and such properties – as opposed to person with such and such proper-
ties – may end up being part of the pre-pre-semantics of “Smith”, at least for me.) Ex-
actly similar remarks apply to “Cicero”, “Tully”, “Hesperus”, and “Phosphorous.” What
Frege thought of as the sense of “Hesperus” is really the pre-pre-semantics of tokens of
that expression.
Frege’s notion of sense is thus a conflation of three different concepts: token-mean-
ing, type-meaning, and the non-semantic information through which one ascertains
type-meaning.
Kaplan (1989b: 568) writes:
Fregean sinn conflates elements of two quite different notions of meaning. One,
which I called character, is close to the idea of linguistic meaning (and perhaps of
cognitive content). Another, which I call content, is what is said by an expression
in a particular context of use. The content of an utterance [i.e. a token] of a com-
plete sentence is a truth-bearing proposition. Where indexicals are involved, the
difference between character and content is quite clear.
 Conceptual Atomism and the Computational Theory of Mind

Kaplan’s point contains a deep insight, and it is correct as far as it goes. But it doesn’t
go far enough. Frege didn’t just hold that demonstratives and definite descriptions
have sense; he didn’t just hold that “non-rigid” designators have sense. He held that
proper nouns, like “Socrates” and “Hesperus”, have sense. But we know from Kripke
that they don’t (at least not at the level of literal meaning), and therefore that the char-
acter-content distinction isn’t applicable to such expressions.
In any case, even if it is applicable, it doesn’t explain why tokens of “Hesperus is
Phosphorous” communicate non-trivial propositions. There is some x such that the se-
mantic content of the expression-type “Hesperus” is tokens of “Hesperus” refer to x and
such that the semantic content of the expression-type “Phosphorous” is tokens of “Phos-
phorous” refer to x. We could say that these rules are the “characters” of those expressions.
But those rules are not sufficiently different to explain why tokens of FHesperus has phiG
convey propositions concerning the concept last celestial body to disappear from the
morning sky or why tokens of FPhosphorous has phiG convey propositions concerning the
concept first celestial body to appear in the evening sky. Those “characters” cannot even
begin to explain the facts relating to the cognitive significances of such tokens.
So while Kaplan is quite right to say that Frege’s notion of sense conflates character
and content – type-meaning and token-meaning – that is not all that it conflates. It
also conflates the information through which one computes type-meaning with type-
meaning. Thus Frege’s analysis involves a conflation, not of two, but of three different
notions: semantics (token-meaning), pre-semantics (type-meaning), and pre-pre-se-
mantics (the information through which one computes type-meaning and, on that
basis, token-meaning).
To illustrate our criticism of Frege’s analysis, we chose to consider to a case where
reference was learned ostensively. Of course, one doesn’t have to learn what “Hespe-
rus” refers to through an ostensive definition; indeed, this isn’t likely to be how one
learns it. But everything that we’ve said about cases of ostensive learning carries over
to cases of non-ostensive learning. As we saw, it is always through a descriptive-exis-
tential claim that one learns what an expression E means; it is through some piece of
information having the form: there is some object x having such and such properties,
and tokens of E refer to x. Given this, everything that we said about cases where one
learns the meaning of “Hesperus” ostensively holds in cases where one learns that
meaning non-ostensively. Of course, everything that we just said about “Hesperus” is
true of any other proper noun.
Semantics is public. The expression-type “Hesperus” has a public meaning, and so
do tokens of that type. But pre-pre-semantics varies from person to person. Depend-
ing on how I learn what “Hesperus” refers to, tokens of “Hesperus” might or might not
connote the concept last celestial body…to me; and in virtue of their containing occur-
rences of “Hesperus”, sentence-tokens might or might not convey propositions con-
taining that concept to me. This is consistent with the fact that semantics is public.
After all, pre-pre-semantics is not semantics. It isn’t token-semantics or type-seman-
tics. It is how people access semantics.
 Chapter 5.  Modality, intensionality, and a posteriori necessity 

Given that pre-pre-semantics can vary from person to person, it might seem that
our analysis threatens the possibility of interpersonal communication. The opposite is
the case. Because of this variability, what one’s words literally mean is not in complete
lock-step with what one is thinking: there is always a certain slack between the two. And
it is because of this slack that one person’s words can be meaningful to another person,
even if the two people have had very different psycho-biographies. As we will discuss, if
the literal meanings of one’s utterances were too close in content to the thoughts that
precipitated them, those utterances would not be intelligible to others. Communication
involves isomorphism, not identity, of the thoughts involved in the production and com-
prehension of linguistic expressions. We will investigate this in Chapter 9.

Extensionality revisited

As we’ve seen, a token of “Hesperus is lovely” may convey a very different proposition
from a token of “Phosphorous is lovely.” Given what Kripke (1972) and Soames (2001,
2005) say, it is probably not an option to hold that those tokens differ in literal mean-
ing. And it isn’t necessary to do so. It is necessary only to distinguish literal meaning
from the information through which it is ascertained. It is necessary only to distin-
guish semantics from pre-semantics (and from pre-pre-semantics).
An obvious extension of this line of thought reinforces the views on extensionality
that we put forth in the previous chapter. A token of “John believes that Hesperus is
lovely” may communicate a proposition that differs in truth-value from the proposi-
tion that is communicated by a token of “John believes that Phosphorous is lovely.”
Given what Kripke and Soames say, it isn’t an option to hold that those tokens differ in
literal meaning. And it isn’t necessary to do so: that difference in truth-value is to be
explained along lines similar to those used to explain the apparent difference in mean-
ing between “Hesperus is lovely” and “Phosphorous is lovely.” Yet another threat to
extensionality turns out to involve a conflation of semantics and pre-semantics (and
pre-pre-semantics).
Given what Kaplan (1989) says, tokens of “he is tall” and “you are tall” have the
same literal meaning if the occurrence of “he” co-refers with the occurrence of “you.”
In Chapter 1, we explained why Kaplan’s view is consistent with the fact that those to-
kens convey very different propositions. A token of “John believes that you are tall”
may convey a proposition that differs in truth-value from the proposition conveyed by
a token of “John believes that he is tall.” But given obvious extensions of what we said
in Chapter 1, that is perfectly consistent with the supposition that those tokens have
the same proposition for their literal meaning. And given what Kaplan (1989) says, it
is probably necessary to suppose that they have the same literal meaning. Yet another
threat to extensionality is neutralized.
There is no shortage of authors who think that truth-value can be changed by in-
tersubstituting co-referring definite descriptions, and who thus think that not all con-
 Conceptual Atomism and the Computational Theory of Mind

texts are extensional.5 The arguments in favor of that view are exact analogues of the
arguments in favor of the view that truth-value can be changed by intersubstituting
two co-referring proper names or demonstratives. The latter arguments are fallacious,
and the same is therefore true of the former. Further, as we saw in Chapter 3, the sup-
position that definite descriptions are directly referential is easily reconciled with the
fact that the truth-value of what is conveyed can be changed by intersubstituting co-
referring definite descriptions. In light of these points, it would be very artificial to
suppose that intersubstituting co-referring definite descriptions changed truth-value.
It thus appears that intersubstituting two co-referring nouns (or noun-phrases) never
changes truth-value and that all contexts are therefore extensional.6

5. See Kuczynski (2006d).

6. Anticipations of much of what we just said are found in Boguslawski (1994).
 chapter 6

Cognitive maps and causal connections

Why the causal story is an important part of
the descriptive story

It has often been said, both by semanticists and philosophers of mind, that one must
be causally connected to an object in order to think about it.1
This view seems to conflict with the fact that we can refer to things to which we
have no causal connection. The famous example is Kaplan’s “Newman 1” (Kaplan
1975), which is defined thus: if there is an x such that x born before anyone else in the
3rd millennium A.D., let “Newman 1” refer to x. So if Brown has that property, then
(N1) “Newman 1 will be a genius”
semantically encodes the proposition:
(BG) Brown will be a genius.
Some authors (Donnellan 1974, Kaplan 1989) hold both that we can refer to things to
which we are not causally connected and that we must be causally connected to a thing
in order to think about it. They therefore hold that we don’t grasp the proposition se-
mantically encoded in (N1). We are not causally connected to Brown, at least not in
the right way. Therefore we cannot think about him, and thus cannot grasp the propo-
sition that is literally meant by (N1). But because we know the relevant semantic rules,
we can still refer to him.
Given our discovery that to have a concept of x is to know a proposition that de-
scribes x, this reasonable-seeming view should be reconsidered. I say “reconsidered”,
not “rejected”, because we will see in this chapter that it is correct and that, despite first
appearances, it is in no way incompatible with the descriptivist conception of concep-
tion that we’ve been advocating.
Let us start by reviewing some fundamental points. In many cases, a person’s abil-
ity to think about a thing involves his having a certain causal relation to that thing. But
we’ve seen that, in every such case, the causal connection between subject and object is
simply an ingredient in the subject’s knowledge of the right kind of uniquely individu-
ating description. Your acquiring a concept of Fred by seeing him consists in your ac-
quiring knowledge of an existence-claim of which Fred is the unique verifier. If you

1. Keith Donnellan (1974) and Jerry Fodor (1990, 1998) hold this view. Kaplan (1968) held it
early in his career, though he later retracted it (1989b).
 Conceptual Atomism and the Computational Theory of Mind

acquire a concept of Fred by being told about him or by hearing about him, that too
consists in your acquiring knowledge of an existence-claim of which Fred is the unique
verifier. In all cases, then, knowledge of a thing is identical with knowledge of a unique-
ly individuating description that applies to (i.e. is satisfied by) that thing. In some cases,
a causal connection between subject and object is an ingredient in that knowledge. But
it is never the causal connection per se that constitutes conception; what does so is al-
ways knowledge of a uniquely individuating description.
Obviously we can know of a uniquely individuating description that applies to
Brown (Newman1). So given only what we said in the last paragraph, it seems wrong,
or at least arbitrary, to say that we can think about Socrates but that we cannot think
about Newman 1. For the very thing that enables us to think about Socrates would
seem to be present in the case of Newman 1.2

2. We have said that to have a conception of x is to know some true proposition of the form…
C…, where C is a uniquely individuating description of x. There are, of course, cases where a
causal connection to x is essentially involved. But, as we have seen, the causal connection is
always embedded in knowledge of a proposition of the kind just described.
But there is an obvious problem with this view: a problem that doesn’t threaten our posi-
tion, but must still be dealt with. I myself do not know who invented the fugue. But let O be that
person, whoever it is. (So, in this context, “O” will be a proper name denoting that person.) I
certainly know some true proposition of the form:…C…, where C is a description that applies
uniquely to O; for I know that the following are true:
(i) Somebody x uniquely invented the fugue and he had at least some musical ability.
(ii) Somebody x uniquely invented the fugue and he lived before the 19th century.
(iii) Somebody x uniquely invented the fugue and x invented the fugue.
Obviously knowing some uniquely individuating description that applies to O isn’t enough to
know who O is. And it therefore isn’t enough to know the identities of the propositions seman-
tically encoded in sentences like:
(iv) “O was a talented composer”.
(v) “O died of scurvy.”
(vi) “O incurred the wrath of his less talented contemporaries.”
What we just said about “O” precisely parallels what we said earlier about “Julius.” The semantics
of the sentence-type “Julius was tall” is given by the rule: If there is some x such that x uniquely
invented the zipper, then a given token of “Julius was tall” means: x was tall. One can know that
rule without knowing who invented the zipper and, therefore, without knowing what is meant
by any given token of that type. (At the same time, a knowledge of what is meant by that sen-
tence-type can simulate a knowledge of what is meant by tokens thereof. As we’ve seen, most of
what is communicated by utterances lies, not in what is literally meant by them, but in what is
learned in the process of figuring out, or attempting to figure out, what is literally meant by
them. Thus, given two people, both of whom know the semantics of the type “Julius was tall” but
only one of whom knows who invented the zipper, what an utterance of that sentence conveys to
the one will ceteris paribus overlap to a high degree with what it conveys to the other. In general,
coincidences in pre-semantic knowledge often counterbalance or even outweigh differences in
semantic knowledge, and pre-semantic knowledge can therefore sometimes mimic non-existent
 Chapter 6.  Cognitive maps and casual connections 

Why the causal story is a necessary addition to the descriptive story

Nonetheless, there is a significant difference between the sort of concept that we have
of Newman 1, on the one hand, and the sort of concept that we have of Socrates or of
those with whom we are personally acquainted. Donnellan (1974) and Evans (1982),
are therefore right to put those concepts in different categories. Donnellan and Evans
are also right to see the relevant difference as lying in the fact that in the one case, but
not the other, a causal connection between subject and object is involved.
A proper understanding of this issue is to be found in an insight of Evans’.3 Evans
(1982: 65) held that, in order for x to be able to have thoughts about y, x must have a
“discriminating conception” of y. Unlike us, Evans did not hold that such knowledge
consists in knowledge of a uniquely individuating description of the kind described
earlier. But, like us, he rejected the idea that, by itself, a causal connection between x
and y could possibly enable x to have thoughts about y. (He referred to the idea that, by
itself, such a connection could realize conception as the “photograph model.”) And,
like us, Evans held that, although such a causal connection has an important role with-
in conception, it must be embedded within knowledge of a certain kind – knowledge
that singles out the relevant object, and thus “discriminates” between it and everything
else – if there is to be any concept.
What, more exactly, is the nature of that knowledge? Evans (1982: 143–191, 1985:
291–321) rightly says that spatiotemporal individuals are distinguished from one an-
other by their locations (in both space and time). Of course, Socrates is not individu-

semantic knowledge. In previous chapters, we found some widely accepted contentions to em-
body a failure to distinguish type-semantic knowledge from bona fide token-semantic knowled-
ge, and in this chapter we will find the same to be true of some doctrines additional to those
already considered.) To put it another way, somebody who doesn’t know what is meant by to-
kens of “Julius was tall” may still know that somebody x uniquely invented the zipper and that
“Julius” refers to x. It follows that knowing to whom “Julius” refers involves than knowing just
any truth of the form: if somebody x uniquely falls under concept C, “Julius” refers to x.
What else is needed? In this chapter, I am going to provide one answer to that question. But
there is an answer different from the one that I am going to provide that strikes me as plausible,
and I would now like to state that other answer. (I will set aside the question whether those
answers are compatible.) To identify is to reidentify. To know who so and so is, one must know
of two existence-claims that are uniquely verified by so and so and, further, one must know that
some one person uniquely verifies them. Suppose you meet Bob on Monday. Since you have just
met him, there is a sense in which you don’t know who he is. If somebody asks you, “do you
know who this is?”, you must say: “I am afraid I do not.” But, of course, in meeting him, you are
uploading information that enables you to think about him – you are acquiring a concept of
him. As we saw, this information is existential. Suppose that, a day later, you see Bob again. This
time, there is a sense in which you know who you are seeing. As we saw earlier, your recognizing
Bob consists in your knowing the truth of two existence-claims of which he is the sole verifier,
and in your also knowing that some one entity verifies both of them.
3. See especially Evans (1985: Chapter 6).
 Conceptual Atomism and the Computational Theory of Mind

ated by the fact that, at some specific moment, he is in some specific place. After all, for
any time t, if Socrates was at place p at t, Socrates might have been somewhere other
than p at t. But every spatiotemporal individual has unique space-time co-ordinates.
For this reason, it seems reasonable to suppose that distinguishing one individual from
another involves one’s having some conception of how those individuals differ in re-
spect of their space-time locations.
Of course, if you see some object, you ipso facto know, relative to an egocentric
frame of reference, how that object is spatiotemporally distinctive. So there is no trou-
ble seeing how x’s perceiving y gives x the sort of discriminating knowledge of y that,
in Evans’ view (and ours), endows x with a concept of y. Obviously seeing an object
involves being causally connected with it. We saw earlier how this causal connection is
a mere constituent of one’s being given a uniquely individuating description of that
object. So if we confine ourselves to cases of sense-perception, it is clear how to recon-
cile the following two views (both of which we and Evans accept): first, in at least some
cases, a causal connection between x and y is constitutive of x’s concept of y; second, x
must have “discriminating knowledge” of y to have a concept of that object.
But, of course, most of our concepts are not acquired directly through perception.
My concept of Socrates didn’t result from a perception of him. At the same time, that
concept, like a visual perception of Socrates, has a causal component. And we have
seen other similarities between conceptions that are derived from perceptual encoun-
ters with objects (e.g. my concept of my best friend) and conceptions are not so de-
rived (e.g. my concept of Socrates). This suggests that it is possible to extend what said
a moment ago about concepts of the first kind to concepts of the second kind. We will
attempt to make that extension in this chapter.
A caveat is in order. It is, I will argue, only where spatiotemporal objects are con-
cerned that a causal connection between subject and object is constitutive of concep-
tion. We will not attempt to show that what is true of my concept of my best friend or
of Socrates is true of my concept of the number two or the concept of justice.4
But supposing that, at least where spatiotemporal objects are concerned, causality
is constitutive of conception, we must ask why that is so. Why is a causal connection
between x and y so important to x’s having a concept of y? Is it just a brute fact? Or, on
the contrary, can we find some basis for that fact?

Causality and spatiotemporal locatedness

I am going to give an Evans-inspired answer to this last question. (But even though it is
Evans-inspired, my answer is not one that Evans himself provides. In fact, given his
belief that our thoughts are Russellian with respect to constituents of the external world,

4. Fodor (1990, 1998) argues that conception always has a causal component. In Chapters 22-
23, we will see argue that this contention is false.
 Chapter 6.  Cognitive maps and casual connections 

it is not one that he could provide.) First of all, spatiotemporal relations are causal rela-
tions. E1 temporally precedes E2 iff there is a possible causal process (e.g. a light-ray) that
begins with E1 and ends with E2 (Reichenbach 1958, Sklar 1974, Salmon 1984). Indeed,
it would make no sense to say that two events were part of the same temporal order un-
less either (i) they were connected by a (possible) causal sequence or (ii) each of them
were connected by such a sequence to some third event. Further, it would make no
sense to say that two events were part of the same space unless either (i*) they were con-
nected by some (possible) causal sequence or (ii*) they were both connected by some
such sequence to some third event (Reichenbach 1958, Sklar 1974, Salmon 1984).
Although recent, and extremely sophisticated, developments in theoretical phys-
ics were needed to force these facts on our consciousness, it is hard to believe that we
have not always known them at some intuitive level. Intuition recoils at the idea of two
events having temporal or spatial relations but being necessarily causally divorced both
from each other and also from any third event. Spatiotemporal relations are insepara-
ble from causal relations. It seems reasonable to believe that, at some cognitive level,
everyone has an awareness of the intimate connections between spatiotemporal loca-
tion and causality. Indeed, it seems reasonable, if not mandatory, to suppose that an
awareness of this constitutes a vital part of our “conceptual scheme” – our way of
processing the disturbances of our sensory surfaces and of subsequently generating
perceptual and cognitive representations of external reality.
Of course, this knowledge is not discursive or even conscious.5 But we wouldn’t
expect it to be conscious, given that this knowledge is constitutive of the apparatus that
generates perceptions and other contents of consciousness. This knowledge is part of
our cognitive architecture. It thus underlies the representations of the external world
that are generated by conscious and, therefore, by discursive thought. For this reason,
that knowledge is not represented within those representations, except in so far as our
thought has reached an extraordinarily high-level of self-awareness.

The egocentric character of cognitive maps

We thus have three facts to work with. First, spatiotemporal relations are causal relations.
Second, at some level, everybody knows this. Third, spatiotemporal individuals are distin-
guished from one another by their spatiotemporal co-ordinates.
Before we can close the argument, there is a fourth point to make. Ultimately, one’s
cognitive map of the world is egocentric. In this respect, cognitive maps are funda-
mentally different from the maps produced by cartographers and astronomers. While

5. Except where a few scholars are concerned. But, even then, what is conscious is not the
knowledge of which I am speaking, and is rather an analogue of that knowledge. The knowledge
acquired by the cognitive scientist doesn’t help generate sense-perceptions, and is therefore qua-
litatively, and thus numerically, distinct from the knowledge of which the cognitive scientist has
just acquired knowledge.
 Conceptual Atomism and the Computational Theory of Mind

there is a sense in which maps of the latter kind are egocentric, it is entirely different
from the sense in which cognitive maps are egocentric.
Right now let us say exactly how the maps produced by professional map-makers
are egocentric. A few pages hence, in the section titled “the concept of a cognitive map”,
we will see why the egocentricity of such maps is of a much more superficial kind than
that characteristic of cognitive maps.
Even if you are an extremely knowledgeable geographer or astronomer, you must
ultimately understand the location of a given country, or a given galaxy, in terms of
something that is given to you perceptually. The difference between somebody who is
knowledgeable about geography and somebody who is not is that the former is able to
put his concept of this place [the place where I, the knowledgeable geographer, am] into
a much wider context than the person who doesn’t know anything about geography.
Nobody understands the layout of Earth’s surface (or of the Solar System or of the Cos-
mos…) in completely non-egocentric terms. The astronomer (or geographer) can inte-
grate his thoughts of the form …this place…into a much wider body of information than
the astronomically (or geographically) uninformed. But, in the end, the astronomer
must begin with some kind of purely perceptual, and therefore egocentric, interface with
his environment. So every cognitive map of the world, no matter how inclusive it ends
up becoming, begins with the egocentric phenomenon of sense-perception. Whenever
one locates an event in space-time, one does so in terms that are ultimately egocentric.
From these points, it follows that an event’s (or object’s) spatiotemporal relations
to a given person are indistinguishable from its causal relations to that person (or, in
any case, from a certain subset of those relations). To locate an object in space-time –
i.e. to establish its spatiotemporal position relative to oneself, is to know of some caus-
al connection connecting that event to one self.
To perceive an object is to have an awareness of a very direct causal link between
that object and oneself. Obviously, considerable scientific sophistication is needed to
know the exact nature of the link. But, as previously discussed, if x perceives y, it is
hard to believe that at every cognitive level x is ignorant of the fact that y is having a
rather direct effect on his own experiences, and it is also hard to believe that x is en-
tirely ignorant of the nature of that causal relation.
Suppose that somebody shines a flashlight in your eyes. Perhaps you know noth-
ing about the physics or neurophysiology involved in the processes mediating between
the shining of the flashlight in your direction and your subsequent flash-blindness. But
you have a great deal of comparative knowledge with regard to this process. First of all,
you know that a causal connection is involved: you know that the unpleasant state you
are in is a consequence of the fact that a beam of light is pointing in your direction.
Second, you can certainly discriminate between that kind of causal process and others.
You know that you are at the receiving end of something that is qualitatively different
from a cool breeze or a jet of scalding water. So, although you may have little or no
theoretical understanding of the mechanics of these various processes, you are able to
distinguish between them. You thus have a great deal of differential knowledge as to
 Chapter 6.  Cognitive maps and casual connections 

the nature of this process, granting that your theoretical or scientific knowledge of it
may be limited or even non-existent.
Similarly, somebody x who sees (or feels or hears…) something y must surely reg-
ister, at some psychological level or other, the fact that y is causally responsible for his
current condition and, given the points just made, must therefore have a reasonably
good differential understanding of the nature of that process.6
It is hard to believe that, when a cat sees a dog charging at him, there is no repre-
sentation at all in the cat’s mind of some kind of causal nexus between the cat’s current
experiences and the occurrence of a certain kind of external phenomenon. Obviously
the representation in question doesn’t consist in a discursive, theoretically articulated
understanding of the sort that only our best scientists can produce. But it would be
absurd to hold, on that basis, that at every cognitive level the cat was altogether igno-
rant of any relations of causal dependence of the sort just discussed. And the behavior
and cognition of animals would be inexplicable unless some allowance were made for
some kind of awareness on their part of relations of causal dependence holding be-
tween the characters of their own psychological states, on the one hand, and those of
events in the external world, on the other.
Given these points, it is not hard to see why perception has such a distinguished
and fundamental role in concept-formation. To have a concept of a thing is to be able
to distinguish it from all other things. Ultimately, spatiotemporal individuals are distin-
guished by their space-time co-ordinates. Spatiotemporal distinctions are causal dis-
tinctions of a certain kind. All knowledge of spatiotemporal distinctions is ultimately
egocentric. To perceive something is to have knowledge of a causal relation between
that thing and oneself, and it is also to have a great deal of (differential, not theoretical
or otherwise absolute) knowledge as to the nature of that causal relation. So it is because
it has a causal component that perception can give us the uniquely individuating knowl-
edge of objects that is identical with our having concepts of them. It is therefore not a
brute fact that perception is, at least in part, a causal phenomenon. That fact can be
understood in terms of the fact that perception gives one an understanding, albeit an
egocentric one, of the spatiotemporal locations of individuals and states of affairs.
I thus propose that for x to have a concept of y is for x to have some idea of y’s
causal relation to onself. (This line of thought is not meant to apply to our knowledge
of abstract objects.) But the reason for this is not that a causal connection between x
and y is a fundamental, inexplicable fact about x’s having a concept of y. What is thus
fundamental is x’s having discriminating knowledge of y. And what is fundamental to
x’s having such a conception is his having (egocentric) knowledge of y’s spatiotemporal
location. Finally, what is fundamental to that sort of knowledge is x’s knowledge of the
causal relation that y bears to him.

6. See Searle (1983: 225-228) for a similar line of thought.

 Conceptual Atomism and the Computational Theory of Mind

Conception of an object as assigning a place to it in one’s cognitive map

Before we can close the argument, there are two more points to make. To have a cognitive
map of the universe is (tautologically) to know how things are spatiotemporally distrib-
uted in it. A corollary is that, in some cognitively pregnant sense of the word “concept”, x
has no concept of a thing y unless x has some way of fitting y into his cognitive map.
So, just as Kaplan (1968) and Dretske (1982) and others say, the difference be-
tween having and lacking a concept can come down to down the presence or absence
of a certain kind of causal connection between subject and object.7
But philosophers of an externalist turn of mind over-react to this. They say, or at least
gravitate towards the position, that concepts are to no degree descriptive and are to be
understood in strictly causal terms (Dretske 1982, Fodor 1990, 1998). They say that con-
cepts are related to their objects in the same way that discolorations on a photographic
plate are related to the events that caused those discolorations. In any case, the latter rela-
tionship is taken as a model of the former. But we’ve seen why, in addition to being radi-
cally counterintuitive, such a view is demonstrably false. (Later, when we discuss concep-
tual atomism, we will encounter additional reasons to reject the photograph-model.)

Putting the causal story into context

Here is the reaction that one ought to have to the fact that a causal connection between
subject and object is constitutive of the former’s concept of the latter. To have a concept of
object x is not just to know the truth of some claim of the form something uniquely has phi
that x satisfies. That is part of the story, but not the whole of it. The existence-claim in
question must be such that in virtue of knowing it, one is able to integrate the verifier of that
claim into one’s cognitive map of the world. That is why, supposing that Smith and Smith
alone invented the zipper, a knowledge of the truth of the proposition somebody x unique-
ly invented the zipper does not give you a concept of Smith. (For it is not the case that, in
virtue of your knowing that proposition, you are able to assign a position to Smith in your
cognitive map.) And that is also why, supposing that Smith is the one person whom you
are now seeing, a knowledge of the truth of the proposition somebody x is uniquely such
that I am currently seeing x does give you a concept of Smith. (For in virtue of your know-
ing that fact, you can assign a place to Smith in your cognitive map.)
In general, for x to have a concept of y is for x to know the truth of some existence-
claim E such that (i) y uniquely satisfies E and such that (ii) in virtue of knowing E, x
is so causally related to y that x can integrate y into his cognitive map. Thus, just as
Dretske and Fodor say, a sine qua non of x’s having a concept of y is that x be at the
receiving end of a causal process beginning with y. But it is important that x satisfy that

7. I repeat that, in this context, “conception” refers only to our conception of spatiotemporal
entitie. Our concepts of abstracta are dealt with in Chapters 22 and 23.
 Chapter 6.  Cognitive maps and casual connections 

condition only because, if he doesn’t, he will lack knowledge of the right kind of de-
scription of y. So, in the end, conception is descriptive, and a causal connection be-
tween subject and object is important only to the extent it allows the subject to have
the right descriptive knowledge.

How this analysis applies to objects that are not directly known
through sense-perception

Of course, there are cases where x’s concept of y is not derived from a perception of y
itself. Consider my concept of Socrates. By itself, that concept gives me no information
about Socrates’ spatiotemporal co-ordinates relative to me. So, by itself, that concept
doesn’t tell me how Socrates fits into my cognitive map.
But this does not ultimately pose any problem for our analysis of conception. Sup-
pose that my concept of Socrates consists in my knowing existence-claim E*. For rea-
sons that we discussed in Chapter 3, Socrates will uniquely satisfy E* and, moreover, E*
will be such that, in virtue of knowing it, I will stand in a specific sort of causal relation
with respect to Socrates. More exactly, I will be at the receiving end of what some phi-
losophers refer to as an “information-transmitting”8 sequence S beginning with Socra-
tes. So even though my knowledge of E* does not by itself enable me to integrate Socra-
tes into my cognitive map, my knowing E* nonetheless gives me the information that I
need to effect that integration.
So, after a fashion, Kaplan (1968), Dretske (1982), and Fodor (1990) are right. To
have a concept of Socrates, I do need to be at the end of a certain kind of causal process
that begins with Socrates. But contrary to what Fodor holds, this is because my con-
cept of Socrates consists in my knowing an existence-claim (of a certain kind) that
describes Socrates.
I have been making liberal use of the term “cognitive map”, but have not yet pro-
vided a definition of it. Let me now do so. The motivation for much of what was just
said will emerge in the process of giving that definition.

The concept of a cognitive map

By a “cognitive map”, I am not referring to a theoretical distillation of scientific findings

concerning the locations of individuals in the space-time manifold. Every creature
having a minimum of cognitive sophistication – every creature capable of perception
and action – is able to orient itself with respect to other individuals in its immediate
environment, and is thus able to represent objects and states of affairs as having loca-

8. Dretske (1982), Fodor (1990).

 Conceptual Atomism and the Computational Theory of Mind

tions, in both space and time, relative to itself.9 This ability is not the result of empirical
inquiry. Rather, it is a pre-condition for such inquiry. Indeed, it is hardly distinguish-
able from a creature’s ability to have the perceptions through which empirical knowl-
edge is acquired. Since the cognitive processes mediating perception are sub-personal,
the same is true of the cognitive map in question (even though, obviously, those sub-
personal processes result in mental representations that belong to the personal stratum
of mentation).
There is a marked difference between, on the one hand, the cognitive maps that
underlie our most basic dealings with the world and, on the other hand, the maps that
cartographers and astronomers produce. Consider a map of the surface of the Earth. If
that map is to give you any information, you must be able to correlate some item on
that map with some location that is given to you directly. You must be able to have a
thought of the form “this place [referring to the place one is occupying] is there on the
map” (Evans 1982: Chapter 6).
In general, to interpret a map, you need some kind of a reference-point in your
own perceptual experience. You thus need to have some independent grasp of at least
one of the places represented by the map. Otherwise the map gives you, at most, the
locations of the items depicted on it with respect to each other, but not with respect to
you. And in that case, of course, the “map” in question isn’t a map at all, at least not
from your viewpoint, since it cannot tell you where you are on the map.
So to understand any map – even one of the kind that cartographers and astrono-
mers produce – you must be able to correlate at least some point on that map with an
indexical or “egocentric” understanding of your own location.
Nonetheless, the maps of cartographers are fundamentally different from your
cognitive map. The difference is that, notwithstanding what we just said, the former are
fundamentally non-egocentric, whereas the latter is fundamentally egocentric. As long
as you can correlate your present location with some point on (say) a map of the Earth’s
surface, you will have no trouble understanding the map, it being irrelevant which
point it happens to be. Suppose you point to the dot that represents Topeka. If, while
pointing to that dot, you can truthfully say “I am there”, then you are in a position to
understand the map. Now replace the occurrence of “Topeka” in the statement just
made with the name of any other place represented on the map. The resulting state-
ment goes through. So while it is true that, in order to understand the map, you must
relate it to a place that is given to you in egocentric terms, the map itself does not have
an egocentric character. The objective, non-egocentric character of the map is ex-
pressed in the fact that any place represented by that map will serve just as well as any
other as a reference-point.
In light of this point, consider some arbitrary map M with the following property.
Suppose that there is some place L such that, in order to understand M, L specifically

9. This is a point that pervades Evans (1982). It is because of a reading of that work that I ap-
preciate its significance. I found Rick Grush’s commentary extremely helpful.
 Chapter 6.  Cognitive maps and casual connections 

must be your reference point. Suppose, for example, that in order to understand M,
you must be able to have a true thought of the form: I am now in Topeka, and that dot
on M represents Topeka. In that case, M distorts the mutual relations of the places de-
picted by it. To that extent some places on a map are better reference-points than oth-
ers – to the extent that certain places on it are more privileged than others – that map
misrepresents the relevant geographical area.
A personal anecdote may help to bring out my meaning. I once owned a coffee-mug
that said on it “A Washington D.C. resident’s map of the world.” Beneath the caption was
a large picture of the Washington Monument, a smaller picture of the Empire state build-
ing, a tiny picture of the Eiffel Tower, a miniscule picture of the Taj Mahal, and so on. Of
course, that map was not a good one. The reason is that, for that map to be at all useful,
there is some one location L on the map such that one must be able to correlate one’s
location with L. That map was able to provide me with geographical information only
because, as a Washington D.C. resident, I was able to correlate my actual location with
some location on the mug – only, in other words, because I was able to have a true
thought of the form: I am now in Washington D.C. and D.C. is represented by that place
on the mug. Had I been in (say) Greenland, that map would have been useless. (More
exactly, it would have been useless except in so far as I could use some second map to
figure out Washington D.C.’s relation to Greenland. But in that case, it would have been
the second map, not the map on the mug, that was giving me information.)
Of course, the maps produced by cartographers and astronomers are not like this.
(At least not in so far as those maps are good ones. To simplify exposition, we will sup-
pose that cartographers and astronomers produce good maps. This supposition is only
approximately true. The map produced by a human astronomer would be to a Martian
what the just-described map-mug would be to a Greenlander.) So the maps that car-
tographers and astronomers produce are, in an important sense, non-egocentric. Giv-
en such a map, there is no one point L on that map such that to understand that map,
one must be able to think truthfully, while pointing to L: I am there. Such a map is
egocentric only in the trivial sense that one must be able to truthfully think: I am there
while pointing to some point or other on that map – any point will do.10
What we just said about a cartographer’s map is true, though usually to a lesser ex-
tent, of the body of geographical (and, more generally, spatiotemporal) knowledge that a
cognitively normal person develops throughout one’s lifetime. One develops a map of the
world, and to a lesser extent of the universe, that is objective, in the sense just described.
But what we just said about a cartographer’s map is not true of the cognitive maps
the underlie perception and action. Those maps are fundamentally egocentric. Where
such maps are concerned, there is only one viable reference-point.
Your current visual perception represents the locations and mutual spatial rela-
tions of various objects. It represents the place that is occupied by the star you are see-
ing, as well as the places occupied by the nearby cactus, horse, and barn. And, so far as

10. See Kaplan (1989) for similar remarks.

 Conceptual Atomism and the Computational Theory of Mind

it is veridical, that perception tells you how those objects are spatially interrelated.
Your perception is an ideographic representation of places, and is thus a kind of map.
But where that visual map is concerned, there is a privileged reference-point. There
is some point p on that map such that ceteris paribus that map’s information about
place x is more accurate than its information about place y exactly if, on the map, x is
closer than y to p. This confirms what we said a moment ago: the cognitive maps em-
bodied in one’s perceptual experience have a fundamentally egocentric character.
Of course, that cognitive map is a precondition for one’s acquiring empirical infor-
mation and thus, in particular, for one’s creating non-egocentric maps of the sort dis-
cussed a moment ago. So the latter, non-egocentric maps have a relatively superficial
place in our dealings with the world, whereas the egocentric maps associated with
perception nearly enough are the cognitive apparatus through which we learn about,
and orient ourselves with respect to, external reality.
Let us deal with an objection to this analysis:
Suppose that Smith is the person who satisfies the claim:

(*) At 3:34 p.m. in the year 2323 A.D, exactly one person will be wearing a hat at
location L,

for some highly specific value of L. In the year 2006, Jones knows that (*) is true
(let us set aside the question how he knows this), but Jones does not know the
truth of any other existence-claim that Smith satisfies.
In virtue of knowing (*), Jones has knowledge that would give him a maxi-
mally precise fix on the spatiotemporal location of some object and that, if the
analysis just proposed is correct, would thus seem to qualify as a concept of Smith
in the most robust sense possible. But under the circumstances, we would say Jones
has no concept of Smith. It would thus appear that your analysis falsely predicts
that, under the circumstances described, Jones does have a concept of Smith.

It is true that, in virtue of knowing (*), Jones has knowledge that would give him a
maximally precise fix on the spatiotemporal location of Smith. But it is relative to the
non-egocentric maps created by cartographers that Jones’ knowledge of (*) would give
him this information. It is not relative to the egocentric map that is fundamental to
Jones’ cognitive dealings with the world. Our analysis says that knowledge of (*) con-
stitutes a concept of Smith only if, in virtue of having that knowledge, one can integrate
Smith into one’s cognitive map. So this objection misses the mark.

Why the causal story, rightly understood, does not help content-externalism

We’ve seen that, in at least some cases, a causal connection between x and y is integral
to x’s concept of y. This might be seen as giving credibility to content-externalism. But
this would not be the right reaction to have.
 Chapter 6.  Cognitive maps and casual connections 

Suppose that, on some occasion, Jones sees Smith and, strictly on that basis, forms
a concept of him. Let C be that concept. C is as clear-cut and unadulterated a case as
there is to be found of a concept’s having a causal component: from a content-external-
ist’s viewpoint, C is a paradigm. But even in a situation where Smith doesn’t exist, Jones
could have a mental state that was identical with C in respect of content and (what fol-
lows) in respect of causal properties. Jones could have C even if Smith didn’t exist.
We’ve already seen why. The content of C does not have the form: Smith has phi. It
has the form there is something x having such and such properties…So the content of C
is not object-involving with respect to Smith. That content is given by an existence-
claim which, under the circumstances, Smith uniquely satisfies. There is thus no rea-
son why C couldn’t exist even if Smith did not. So even though, in virtue of having C,
Jones is causally connected to Smith in a very specific way – in exactly the kind of way
that, according to content-externalism, one must be connected to a thing to have a
concept of it – C consists in Jones’ having knowledge of a description that Smith
uniquely satisfies and that is not object-involving with respect to Smith. Of course, the
kind of concept that Jones has of Smith is, for the content-externalist, a paradigm-case
of a concept. So the falsity of content-externalism is evident even when we consider
the concepts that best conform to its analysis.
 chapter 7

Conception as knowledge of series

of interlocking existence-claims

Given the points made in the previous chapter, our analysis of conception must be
refined if it is to be accurate. Let us start by revisiting some of the points that we made
in Chapter 3 regarding conception. I see Fred at a party. I have never seen Fred before.
My forming a concept of him consists in my learning the truth of some existence-
claim, for example:
(1) Somebody x is uniquely such that, right now, x is standing over there, next to
that fireplace…
Given the passage of time, (1) must be replaced with another existence-claim, for ex-
ample:
(2) Somebody x is uniquely such that, a few hours ago, at a certain moment in the
course of a certain party, x was standing next to a certain fire-place...
Let us suppose that, after a few more days go by, I see Fred again. For reasons that we
discussed in Chapter 3, my recognizing Fred will consist in my learning the truth of
some existence-claim that interlocks with (2) in the appropriate way, for example:
(3) Somebody x is uniquely such that, several days ago, at a certain moment in the
course of a certain party, x was standing next to a certain fireplace; moreover, I
can now see x, and x is wearing a plaid sportcoat and, judging by his body-move-
ments and general countenance, seems to be in the midst of telling a joke.
Of course, what we said about (1) is true mutatis mutandis of (3). The latter must be re-
placed with an existence-claim that appropriately reflects the passage of time, for example:
(4) Somebody x is uniquely such that, several days ago, at a certain moment in the
course of a certain party, x was standing next to a certain fireplace; moreover,
two days ago, on a certain street corner, at a certain time, I saw x, and x was
wearing a plaid sport coat and, judging by his body-movements and general
countenance, he seemed to be in the midst of telling a joke.
 Conceptual Atomism and the Computational Theory of Mind

Now that I know (4), I can, without forfeiting my ability to think about Fred, drop my
knowledge of the existence-claims that initiated my concept of him. Thus, (4) can be
replaced with (for example):
(5) Somebody x is uniquely such that, two days ago, I saw x on a certain street-
corner, wearing a loud sport coat, and (so far as I could tell) gesticulating in a
manner characteristic of someone who is telling a joke...
Now that I can access Fred through (5), I can drop my knowledge of (1)–(4).
But at this point we must go beyond what we previously said about conception.
We don’t want to say that I used to have one concept of Fred and that I now have a
different one. Thus, we don’t want to say that my concept of Fred used to be identical
with a knowledge of (1) and is now identical with a knowledge of (5). Such a view sim-
ply wouldn’t be true to our way of speaking. It would be like saying that I am not liter-
ally identical with the person I was yesterday, given that I have changed my clothes in
the meantime. The concept I now have of Fred is numerically identical with the concept
I had of him a few days ago. So, strictly speaking, a concept is not identical with knowl-
edge of a uniquely individuating description of the sort previously described.
Given these points, one might be tempted to say the following: one has a concept
of x if there is some series of existence-claims that satisfies the following four condi-
tions. Each installment in that series (with the exception of the first one) interlocks, in
the way just described, with its predecessor. Second, x uniquely satisfies each of those
existence-claims. Third, one is able, during at least some part of one’s life, to have
thoughts about x. Fourth, or any time t falling into that time-period, it is because one
knows at least one of the existence-claims just described that one is able, at t, to have
thoughts about x.
Such a position would be almost correct – but not quite. As we saw in the last
chapter, each of those existence-claims must be such that, in virtue of my knowing it, I
am able to assign a position to x in my cognitive map.
To have a concept of x, I do not at any given time need to know every member of
the series just described. It is enough if, at any given point, I know just one of them. As
we saw, I can forget (1)–(4) and still continue to have a concept of Fred. One doesn’t
usually remember the existence-claim that started up a given concept that one has, and
one usually doesn’t remember most of the existence-claims that followed that start-up
existence-claim.
Having a concept of x consists in having “discriminating knowledge” of x, as Evans
puts it. At any given point, such knowledge consists in knowing the truth of some ex-
istence-claim that x uniquely satisfies. But one’s knowledge of any one such existence-
claim can, and probably will, be replaced with knowledge of another, interlocking ex-
istence-claim. Further, each of those existence-claims does not merely single out x.
Each singles out x in a very special way: in virtue of knowing any one of those claims,
one is able to integrate x into one’s cognitive map.
 Chapter 7.  Conception as knowledge of series of interlocking existence-claims 

Frege’s paradox revisited

Let (1)–(5) be defined as in the previous section. Further, let K1 be my knowledge of (1);
let K2 be my knowledge of (2); and so on. K1…Kn is thus the series of pieces of knowl-
edge that constitutes my ability to think about Fred. We might therefore describe K1…
Kn as a “knowledge-series.” Let us now suppose that, one day, I meet a person who is
obviously wearing a disguise. This person is in fact Fred, but I don’t know this. For
many years hence, I keep on bumping into this disguised person. In fact, I form a
friendship with him. I refer to him as “Barney.” (During the entirety of my relationship
with him, Barney never removes his disguise when he is with me.) Let K*1…K*n be the
knowledge-series in question. So K*1 is my knowledge of some claim like:
(1*) somebody x is uniquely such that x is a person standing over there wearing a
strange disguise...
And K*2 is my knowledge of some claim that appropriately interlocks with K*1, for
example:
(2*) somebody x is uniquely such that, a few weeks ago, I saw x standing in a cer-
tain place wearing a strange disguise...
So in this case, there is some x (namely Fred), such that two different knowledge-
streams enable me to think about x. That explains why, under the circumstances, I have
two different concepts of Fred.
Of course, these two streams could converge. This would happen if I learned the
truth of some existence-claim Kn+1 that interlocked with both Kn and K*n. Suppose
that, during my last encounter with “Barney”, he was given to me through the claim:
K*n: there is some man x such that x uniquely has such and such characteristics (e.g. x is
uniquely a friend of mine who goes by the name “Barney”) and x is now standing right
in front of me, and is peeling off a mask....

And suppose that, during my last encounter with “Fred”, he was given to me through
the claim:
Kn: There is some man y such that I’ve known y for several years and such that y is
standing in front of me and y has just peeled off the odd mask he’s always worn
over the years; moreover, y looks exactly like the person I’ve referred to as “Fred”
all these years and (because I know on independent grounds that nobody other
than Fred looks anything like him) therefore is that person.

In that case, Kn+1 will be along the lines of:

Kn+1: There is some z such that z is standing in front of me and such that z is the
person I’ve known as “Fred” all these years, and such that z is also the person I’ve
known as “Barney” all these years.
 Conceptual Atomism and the Computational Theory of Mind

So the two knowledge-streams that we’ve been discussing – namely, K1…Kn and K*1…
K*n – converge in Kn+1. Recognition of that truth that would be expressed by a state-
ment like “Fred=Barney”.
For reasons discussed earlier, the sentences “Fred=Barney” and “Hesperus=Phos-
phorous” have for the literal meanings bare and trivial propositions of the form x=x. But
here we are talking about the recognitions that would be associated, albeit non-seman-
tically, with such utterances. We’ve already explained the nature of that association.

Conception versus identification

In order to be able to think about x, one must be able to cognitively single x out, and this
involves having “discriminating knowledge” of x, to borrow Evans’ expression. But hav-
ing such knowledge is different from knowing x’s identity. I can have a concept of x
without knowing who or what x is. (Evans himself made this point clearly.) Let us as-
similate that point into our analysis.
Suppose that, for five years, I live next-door to a man whom I know as “Bob.” I am
on good terms with Bob, and we occasionally have an amicable cup of coffee together.
During such get-togethers, we exchange no substantive information, and I thus learn
little about Bob.
One day I am at the beach with a friend, and I see Bob. Pointing to Bob, my com-
panion asks me “do you know who that person is?” In this context, the answer is “yes.”
I recognize the man at the beach as my next door neighbor.
But a few weeks later, two stern-faced F.B.I. agents come to my door. They aggres-
sively tell me that “Bob” is the head of the world’s largest drug-cartel and that he has
also been selling weapons to various para-military organizations around the world. Of
course, I didn’t know any of this, and I want to make it clear to the agents that I had no
part in Bob’s malfeasance. So I say “I had no idea who he was.”
In saying this, I am telling the truth. I really had no idea who my next door neigh-
bor was; I had no idea that he was such a evil-doer. So there is some x such that I am
truthfully telling the F.B.I. agents that I do not know who x is. But how can this be?
After all, when I was at the beach I truthfully said, pointing to Bob, “yes, I know who
that is.” So it would seem to follow that there is some object x such that I both do, and
do not, know who x is. A paradox.
But there is no paradox. There is some x such that what I am telling the agents is
that I had no idea that x was involved in various heinous enterprises. So there is some
x such that I am telling the agents that I have no idea that x satisfies the description
person who heads the world’s largest drug-cartel and x gives weapons to para-military
organizations… And there is some man such that I am telling my companion at the
beach that I know that he satisfies some very different description, some description
like: person who lives in the house to the immediate left of my house…
 Chapter 7.  Conception as knowledge of series of interlocking existence-claims 

The sentence FI know who x isG can be true in one context and false in another.
This is because, in any context where it is appropriate to ask Fdo you know who x is?G,
there is some description D such that, in that context, FI know who x isG means I know
that x uniquely satisfies D and because, given two different contexts, the description
that is relevant to the one context may be different from the description that is relevant
to the other. When I am talking with the agents, the relevant description is unique head
of the world’s largest drug-cartel…When I am talking with my companion at the beach,
the relevant description is person who lives in the house to the left of mine… The con-
cept of knowing-who (or knowing-what) is contextual.
But this brings us to a fundamental point. In each of the cases just discussed, my
knowing who Bob is presupposes my having some concept of him. A precondition for
my recognizing the man at the beach is that I already have a concept of him. (You can-
not recognize what you do not already cognize.) I can fail to know of my kindly neigh-
bor that he is a crime lord only because I can already have thoughts about him. So
being able to have thoughts about x is different from knowing who or what x is, even
after we allow for the contextual nature of the latter relation. Thus the discriminative
knowledge discussed a moment ago – the knowledge needed to grasp singular propo-
sitions of the form x has phi – is prior to a knowledge of x’s identity.
 chapter 8

The problem of de re senses

Now we must broach a nest of issues relating to Gareth Evans’ important work on
conception. Up to a point, Evans’ system coincides with ours. Like us, Evans is careful
to distinguish semantics from epistemology. Like us, Evans is a direct reference theo-
rist. (So, in his view, there is some x such that “Socrates was bald” encodes the singular
proposition x was bald.) Like us, Evans believes that, at the level of semantics, “Socra-
tes” is completely non-descriptive and is a mere label.
Like us, Evans is very aware that this singular proposition cannot be grasped in a
vacuum – that it must be grasped through other information. He is aware this interme-
diary information is replete with descriptive content. Evans’ system thus coincides
with ours in some important respects.
There are other important similarities between our system and Evans’. Like us,
Evans rejects the idea that conception can be understood in strictly causal terms. So he
rejects the idea, advocated by Fodor, that concepts are related to their objects in the
same way that the discolorations on a photographic plate are related to the events that
caused those discolorations. He thus rejects what he aptly refers to as “the photograph
model” of conception.1

1. His reasons for rejecting the photograph model are very different from ours. Evans seems
to hold that, if the photograph model were correct, then one could exercise a given concept only
in the context of a belief – and not in the context of a non-affirmative attitude towards a propo-
sition (e.g. an attitude of questioning or supposing). See (Evans 1982: Chapter 3) and Grush
(1999). The idea seems to be that, if my mental representation of a state of affairs just is a causal
connection with that state of affairs, then I must always be mentally affirming the existence of
that state of affairs, so long as my concept of it remains in existence. So far as it can express any
kind of propositional attitude towards what it depicts, a photograph of a man on a horse can
only affirm that there is a man on a horse. It cannot express a conditional proposition or a ques-
tion or a supposition.
This idea connects with independently corroborated and significant lines of thought. Freud
(1965: 24) argued that visual imagery is incapable of expressing propositional attitudes other
than mere affirmation:
“The latent dream-thoughts are…transformed into a collection of sensory images and visual
scenes. It is as they travel on this course that what seems to us so novel and so strange occurs to
them. All the linguistic instruments by which we express the subtler relations of thought – the
conjunctions and prepositions, the changes in declension and conjugation – are dropped, be-
cause there are no means of representing them; just as in a primitive language without any
grammar, only the raw material of thought is expressed and abstract terms are taken back to the
concrete ones that are at their basis.”
 Conceptual Atomism and the Computational Theory of Mind

Like us, Evans believes that for x to have a concept of y (where y is a spatiotempo-
ral individual) is for x to have knowledge of a certain kind. In Evans’ view, x must have
“uniquely discriminating knowledge” of y. This means that x must know of at least one
respect in which y is different from all other individuals. (Of course, it would seem that
one would already have to have a concept of y in order to know of some respect in
which y differed from everything else. I leave it open whether this circularity attaches
to Evan’s own view or only to an imperfection in my exposition of it.)
But, at this point, Evans’ system diverges from ours in a number of important re-
spects. First, according to Evans, it is not generally true that x’s having a concept of y
consists in x’s having knowledge of an existence-claim that is uniquely satisfied by y.
Evans concedes that, in some case, this is what conception consists in. But he does not
think that it always, or even typically, consists in this.

Because dreams are ideographic, as opposed to linguistic, representations, non-affirmative

attitudes towards dreamt contents have to be expressed indirectly. For example, an attitude of in-
credulity (“that would never happen: Lucille would never marry Bob”) has to be expressed by in
some roundabout way (e.g. through an image of there being flying pigs or pink elephants at the
marriage-ceremony). See Freud (1998: 346-366) for an extremely sophisticated discussion of the
difficulties that are involved in giving a strictly pictorial representation to non-affirmative propo-
sitional attitudes. (This work of Freud’s was originally published in 1900.)
More direct confirmation for Evans’ view is found in Rescorla (2003). Rescorla points out
that you can’t form a “conditional map” by juxtaposing two maps. A juxtaposition of two maps
doesn’t result in a single, complex representation: all that is left are two semantically disconnec-
ted maps. So there is, as Rescorla says, no map-negation or map-conditionalization.
Evans’ criticism thus seems to be a kind of extension of a criticism that, in private conversa-
tion with David Pears, I made of the “picture-theory” of thought advocated in the Tractatus. If a
thought is just a picture or isomorph of the corresponding fact (or proposition), then how can one
have conditional thoughts or negative thoughts? A purely pictorial language is thus a non-starter.
(Egyptian hieroglyphs are not mere sequences of pictures; they are grammatically structured.)
So Evans’ point has substance, as it indicates a fundamental fact about ideation. Purely vi-
sual – and, more generally, iconic – representation is fundamentally different from linguistic
representation. Language can express a variety of different attitudes towards a different content,
whereas iconic representation cannot.
But Evans doesn’t put his finger on what is fundamentally wrong with the photograph-mo-
del. What is fundamentally wrong with it is that it doesn’t account for any kind of conception.
Given only that x is a photograph of y, it doesn’t follow that x is some kind of awareness of y – it
no more follows that x is an affirmation of y’s existence than it does that x is a denial of y’s exis-
tence. The problem with the photograph model, as I discuss in Chapter 13, is that it falsely
identifies the permissibility of an inference with the actual making of an inference. Given known
causal laws, x’s state (its being discolored in a certain way) warrants the inference that y occur-
red. Given a film of Smith committing the murder, it can reasonably (though defeasibly) be in-
ferred that Smith committed the murder. But it doesn’t follow that the film is an awareness of the
occurrence of some event involving Smith – unless one has an independent grounds for the hi-
ghly implausible view that a causal inference has been made wherever it can be made.
 Chapter 8.  The problem of de re senses 

Evans thus agrees with us that x’s having a concept of y consists in x’s being able to
cognitively distinguish y from all other individuals. But, unlike us, Evans thinks that
this ability does not, at least not as a rule, consist in x’s knowing some existence-claim
that y uniquely satisfies.

Evans’ argument

Why does Evans think that, in general, a concept of y does not consist of knowledge of
an existence-claim that y uniquely satisfies?
First of all, Evans believes that demonstrative identification is not a kind of de-
scriptive identification. His reasoning is best introduced through a bit of fiction. I am
looking at Smith, and I am not looking at any other person. Prior to this, I had no
concept of Smith. I am thus acquiring a concept of Smith from this perception. At this
time, Smith is the sole person who is.00000565 light-years from me. Even though
Smith satisfies the existence-claim there some somebody x such that, at time t, x
is.00000565 light-years from JM, I obviously don’t think of Smith in those terms – noth-
ing having to do with light-years is any part of my concept of Smith.
Of course, Smith’s being that distance from me is an important fact about my con-
cept of Smith; it is an important fact about my acquisition of that concept. But that fact
isn’t represented within my concept.
So, Evans concludes, when we distinguish facts about our concepts from facts
about what is represented within them, we see that there is no warrant for a descriptiv-
ist analysis of concepts acquired through demonstrative identification. 2

Evaluating Evans’ argument

Obviously the basic distinction Evans is making is correct. Not every fact about a concept
is represented within that concept. (My having a concept of Napoleon involves many facts
involving electrons. But the concept electron is no part of the content of that concept.)
But Evans’ application of this truth seems incorrect. Look at some object – at (say)
some tree in your vicinity. So far as it can be expressed in language, the content of that
perception is not, as we’ve seen, given by any noun-phrase; it isn’t given by “a tree” or
“that tree” or anything of the sort. So far as it can be put in words, that content is given
by a sentence, and that sentence is existential in character.
We have seen that, in general, our perceptions make us aware of objects by appris-
ing us of existence-claims that those objects satisfy. Given this fact, Evans’ argument

2. Or, rather, in my reconstruction of that argument. Evans’ own version of the argument is
quite compressed, so I thought it appropriate to expand it into a number of different steps.
 Conceptual Atomism and the Computational Theory of Mind

immediately implodes. As we will now see, this fact also undermines some important
components of Evans’ analysis of conception.

Evans-style pseudo-thoughts

Evans says that demonstrative-thoughts – the thoughts that prompt one to produce
sentence-tokens like “that tree over there is lovely” and “you are a scoundrel” – are
“Russellian” with respect to the objects that are demonstratively identified. (In this
context, “thought” means “cognitive episode”, not “proposition.”)
What does this mean? It means that the content of the thought motivating a sincere
and linguistically competent utterance of “you are a scoundrel” that is addressed to
Smith has Smith as a constituent, the same being true mutatis mutandis of the thoughts
behind “that tree is lovely” and all other non-empty utterances containing demonstra-
tive expressions. (By a “non-empty” utterance, I mean one that doesn’t contain any
terms that are supposed to have a referent but don’t in fact have one, e.g. an occurrence
of “that man over there” in a context where there is no man in the place in question.)
Here it is important to make a distinction. We have argued that Smith himself is a
constituent of the literal meaning of an utterance on your part of “you are a scoundrel” that
is addressed to him. Evans agrees with us about that. But we have also argued that Smith is
not a constituent of the content of any thought that you have. Evans sharply disagrees with
us about that. He is saying that, under the circumstances just described, Smith himself is a
constituent of the content of a thought of yours. On the basis of this view, Evans puts forth
an extraordinary and, I will argue, false thesis about mental representation.
Suppose that you are looking at Smith and are having the kind of thought that,
under those circumstances, would prompt a sincere and competent speaker of English
to say :“that guy is tall.” Let T be that thought, and let S be the situation just described.
(T is not meant to be the proposition that is semantically encoded in an utterance of the
kind just described. T is the thought behind such an utterance.)
Now consider a hypothetical situation S* that is qualitatively just like S except that,
in S*, you are hallucinating, and there is no man in the place at which you are looking.
So, in S*, you aren’t seeing Smith. Apart from some air molecules, the relevant part of
space is completely empty.
According to Evans, in S*, you don’t have T. So, in S*, your mental state (leaving aside
facts about its external causes) is just like your mental state in S, but in S you are having T,
whereas in S* are not. Therefore, T is “Russellian” with respect to Smith: in any scenario
where Smith doesn’t exist, T doesn’t either (Evans 1982: Chapter 6). Of course, in saying
this, Evans is just being a garden-variety content-externalist. But he goes a step further.
According to Evans, there is a sense in which, in S*, you are having no thought at
all. In S*, there is a kind of cognitive blank where, in S, T would be (Evans 1982: Chap-
ter 6). So in S*, you wrongly think that you are having a certain thought that you simply
aren’t having.
 Chapter 8.  The problem of de re senses 

This claim of Evans’ must be qualified if it is to have any plausibility. There is no

denying that, in S*, you are having some kind of thought. You are having a visual experi-
ence that tells you, albeit falsely, that there is a man in a certain place. If you didn’t have
such an experience, then you wouldn’t be hallucinating; if a person doesn’t have percep-
tual experiences that tell him false things about the world around him, then he is not
hallucinating at all. So if the claim of Evans’ that we identified in the last paragraph is to
avoid being straightforwardly false, we must hedge it. What we must say is as follows:
(EVS) In S*, in having your (hallucinatory) visual experience, you are having many
thoughts. But, in S*, there is a kind of cognitive place, so to speak, where there is a
blank or gap in your thoughts – a cognitive place that, in S, is filled with Smith.
In fact, (EVS) is Evans’ own position. But (EVS) is false. We’ve seen the reasons for this
many times, and we don’t need to repeat them.
Given any situation where Evans would contend that a given thought is Russellian with
respect to external object x, an argument exactly similar to the one just given shows that
contention to be erroneous. We may conclude that, contrary to what Evans says, demon-
strative thoughts are not Russellian, even though demonstrative utterances are Russellian.

Evans-style Russellian thoughts

Let us now discuss a corollary of Evans’ view. Let S** be a situation that is just like S
except that, in S**, it is Brown you are seeing, not Smith. (So, in S**, you are not hal-
lucinating.) Of course, given how we’ve set things up, it follows that, in S**, Brown
appears to you in exactly the way in which, in S, Smith appears to you.
Here is Evans* view. There is some thought T** such that, in S**, you are having
T** and such that, in S, you are not having T**, and such that T** is Russellian with
respect to Brown.
For reasons exactly similar to those given in connection with S*, Evans’ view here
is false. There is no thought that you are having in S** that you are not also having in-
each of S and S*.
Why does Evans believe that demonstrative thoughts are Russellian? His argu-
ment is simple. Suppose you point to Smith and say “that guy is tall.” Let t be this ut-
terance. Let us now state Evans’ premises.
Premise 1: For various reasons, we know that the proposition semantically en-
coded in t is simply: Smith is tall.

Premise 2: Obviously people can understand t.

Premise 3: Understanding t consists in knowing which proposition it has for its

literal meaning. Knowing this involves grasping that proposition. So there is some
proposition P such that you understand t only if you grasp P.
 Conceptual Atomism and the Computational Theory of Mind

Premise 4: P is the proposition that one has in mind when one is looking at Smith
and thinking of him that he is tall.

Two things follow from these premises. First, Smith is a constituent of P; in other
words, P is object-involving with respect to Smith. Second, the demonstrative-thought
behind t is Russellian with respect to Smith. Of course, what we said of that thought is
true mutatis mutandis of every demonstrative thought. In the foregoing argument, t
may be replaced with any token t* of a demonstrative-utterance, so long as the demon-
strative in t* is not empty. We may conclude that demonstrative-thoughts (the thoughts
that lie behind sincere, linguistically competent utterances of sentences containing
non-empty occurrences of demonstratives) are categorically Russellian with respect to
constituents of the external world.
Let us evaluate this argument. First of all, it is valid. Given premises 1–4, Evans’
conclusion follows. Further, the first two premises are obviously correct. The problem
lies in the other two premises. Premise 4 is false. Premise 3 is ambiguous, and the argu-
ment fails on the relevant disambiguation.
We’ve seen why premise 4 is false. Let P* be the proposition that lies behind your
utterance of t. So P* is a proposition that you grasp, and your grasp of P* is what mo-
tivates you to utter t. We’ve seen that P* is not a singular proposition of the form x is
tall and that P* doesn’t have Smith as a constituent. P* is some existential proposition
that Smith satisfies, but of which he is not a constituent.
There is undoubtedly some sense in which premise 3 is true. But that premise is
ambiguous. The reason is that, given any utterance, there are two different things it could
mean to say that so and so “knows which” proposition is meant by that utterance. This
ambiguity corresponds to the distinction between knowledge by acquaintance and
knowledge by description. Understanding a sentence-token always involves having some
kind of knowledge of the proposition which it encodes. But, more often than not, that
knowledge is knowledge by description, not knowledge by acquaintance.
Let Green be an arbitrary person. We’ve seen, time and time again, that if O is a
constituent of the external world, Green can’t just think O. He can think about O only
by way of his knowledge that some existence-claim E is uniquely satisfied. (Of course,
O must be the thing that satisfies E.) In Green’s case, let something x has psi be the rel-
evant existence-claim. So Green grasps O by knowing that something uniquely satis-
fies the claim something x has psi. In other words, Green grasps O by knowing the truth
of the claim: something x uniquely has psi.
So far as Green can grasp the proposition O has phi, his grasping that proposition
involves his grasping an existence-claim. This clearly follows from what we was said in
the last paragraph. It also clearly follows that this claim will be:
(#) Something x uniquely has psi, and x also has phi.
Thus, so far as Green grasps the singular proposition O has phi, he grasps it by grasping
some non-singular proposition – some existence-claim that doesn’t presuppose O’s
 Chapter 8.  The problem of de re senses 

existence. At the same time, there is a clear sense in which (#) singles out or describes
the relevant proposition. So there is a sense in which somebody who knows the truth
of (#) grasps O has phi.
Given any true existence-claim, the truth of that claim supervenes on the truth of
some singular proposition. We might describe the latter as the “truth-maker” of the
former. O has phi is the unique truth-maker of (#). So somebody who recognizes the
truth of (#) is thereby grasping, and also recognizing the truth of, O has phi. The sense
in which such a person recognizes the truth of O has phi is similar to the sense in
which one recognizes the truth of some proposition about Socrates if one believes the
proposition: somebody or other was a greatest philosopher to die of hemlock poisoning,
and any such person believed in the existence of abstract objects. So there is a clear sense
in which one can believe a proposition to be true without grasping exactly that propo-
sition. Grasping a proposition involves singling it out. If one knows a proposition by
description, one has singled it out. Knowledge by acquaintance isn’t necessary, and
often isn’t possible.
Let t# be an utterance of “that thing is green”, where t# is accompanied by an osten-
sion of O. The proposition semantically encoded in t# is O is green. It is a datum that
people can understand t#. And it is a datum that this consists, at least in part, in asso-
ciating that utterance with the right proposition. (So Green understands t# only be-
cause he can associate it with the right proposition.) Evans is right about all of this.
But Evans is wrong to think that understanding t# consists in being acquainted
with O is green and then associating it with t#. It is not possible to be acquainted with
that proposition. It can only be known by description.
Here we need to head off a possible confusion. We saw earlier (in Chapter 3) that there
is a descriptive proposition P such that one can be acquainted with P such that P is true iff
O is green is true. Given this, it might seem that one can be acquainted with propositions
that are object-involving with respect to O. But, as we saw, P is not such that its existence
presupposes that of O; and, consequently, it would be possible to grasp P even if O didn’t
exist. So while it is true that understanding t# consists in associating it with its meaning,
and while it is true that this meaning is O has phi, one grasps this meaning, and effects this
association, by way of grasping some existence-claim that does not have O as a constituent.
Evans’ argument is therefore fallacious.

Evans-style cognitive hallucinations

A related position of Evans’ is similarly fallacious. Let W* be a world where Descartes

never existed, but that is otherwise just like our world. So your condition in W* at time
t is identical with your condition in our world at t, at least so far as this possible given
that Descartes never existed in W*. Now consider any one of the various beliefs that
you have in this world about Descartes – e.g. your belief that he was French. Evans’
holds that, in W*, you simply cannot have that thought. So far as, in this world, your
 Conceptual Atomism and the Computational Theory of Mind

thoughts concern Descartes, you simply fail to have thoughts in W*. So far as a mental
event of yours in this world concerns Descartes, the corresponding mental event in W*
fails to constitute a thought.
Of course, from a psychologist’s standpoint there would be no difference between
those events (unless you implausibly hold that, in order to psychoanalyze someone,
you need to know facts about early modern French history). Whatever intelligence,
and whatever neuroses or psychoses, were implicated in the one case would be equally
implicated in the other. But, according to Evans, one of those mental events would be
a thought, whereas the other would not.
Evans would not disagree with our contention that, from a psychologist’s view-
point, your situation in W* would not (in virtue of Descartes’ non-existence there) be
different from your situation here.3 His position seems to be that the category thought
is not a psychological category. (This brings us to a fundamental fact about content-
externalism: it makes facts about representational content be causally, and therefore
psychologically, irrelevant.4 We will explore this in the next section.)
Why does Evans hold that, for each Descartes-thought that you have in this world,
you are shooting a blank – failing to think anything – in W*, even though your psycho-
logical condition here is identical with your psychological condition there?
Evans’ argument for this is similar to the argument that he gives in connection
with demonstrative-thoughts (this is a summary, not a quotation):
Let t be an utterance of “Descartes was French.” We know that, for some x, t’s lit-
eral meaning is the singular proposition x was French. Obviously people can un-
derstand t. Understanding t consists in associating it with the proposition that it
means. Making this association involves grasping that proposition, i.e. in grasping
x was French, for the appropriate x. A person who lives in a world where x exists,
and who grasps the singular proposition x was French, may be psychologically
identical with somebody who lives in a world where x doesn’t exist and who, con-
sequently, doesn’t grasp that proposition. So such people are psychologically iden-
tical, even though one has a thought that the other does not. Counterintuitive
though this conclusion may be, there is no way to circumvent it, given what it is to
understand utterances like “Descartes was French.”

The truth is that there is a way to circumvent that counterintuitive and (so I have argued)
incoherent conclusion. As we saw in Chapter 3, one thinks about Descartes through the

3. It is unclear what McDowell would say here. Given what he says in McDowell (1998: Chap-
ter 13), it seems as though his position would be that there is no significant psychological pro-
perty that you and twin-you have in common in virtue of being molecular twins. It seems, then,
that McDowell would reject any psychological theory – any attempt to formulate psychological
laws – that assumed otherwise. So he would say either that psychology is not a legitimate domain
of research or that psychology must be done along externalist lines – I must know whether the
transparent liquid that we drink is H2O or XYZ or ABC…Both options are highly revisionist.
4. Stich (1978, 1983) makes this very point.
 Chapter 8.  The problem of de re senses 

right kind of uniquely individuating description. And, as we also saw, this is perfectly
consistent with direct-reference-theory, contrary to what has been widely assumed.5

Psychology versus epistemology

Our discussion of Evans brings to light an incoherence that lies at the center of con-
tent-externalism.
Let B be the brain-state or pattern of neural stimulation that realizes your belief that
13 is a prime number. In virtue of being such a belief, B has certain causal powers. For
example, it has the power to generate a belief on your part that there is a prime number
greater than ten. It is thus a matter of empirical fact that, in virtue of its having a certain
content and thus in virtue of its being a thought of a certain kind, B has certain psycho-
logical properties that it would not otherwise have. Given this, it is not an option to deny
that, in virtue of being a thought, B is a certain kind of component of one’s psyche.
It would be deeply implausible to say that my belief that 13 is prime is object-de-
pendent with respect to some constituent of the external world. Surely a brain in a vat
that was just like mine would be capable of having arithmetical thoughts. Surely an en-
vatted brain that was an atom-for-atom duplicate of Gauss’s brain would have at least
some mathematical intelligence, and would therefore be able to grasp mathematical
truths. There thus doesn’t seem to be any possibility of there being a pseudo-thought that
was psychologically just like my actual thought that 13 is prime; and this, for the reasons
given a moment ago, means that content is causally potent, at least in some cases.
But, as Stich (1978) says, it seems deeply arbitrary to suppose that content some-
times has causal powers and sometimes doesn’t. That would be like saying that water is
sometimes H2O and sometimes isn’t. Content-externalism is guilty of exactly this form
of arbitrariness, given that content-externalism sometimes robs content of causal po-
tency, and given that, undeniably, content at least sometimes has causal powers.
A content-externalist would, I believe, be forced to give the following response to this:
You say that content is always potent if it is ever potent. But this is far from self-
evident. Here is what I would propose, in light of the facts that you cited a moment
ago. It is sometimes, but not always, the case that in virtue of having representa-
tional content, something x has causal powers that it wouldn’t otherwise have. In
virtue of being a belief that 13 is prime, something x falls into a certain psycho-
logical category. But it is not the case that, in virtue of being a belief that Alpha
Centauri is lovely, something x falls into a psychological category.
The motivation for this view would be as follows. Let t be some Alpha Cen-
tauri-thought of yours. (So t is object-involving with respect to Alpha Centauri.)
No thought with t’s exact content can occur in a world Alpha Centauri doesn’t

5. Soames (2001, 2005) discusses why, according to many philosophers, there is an inconsis-
tency here, and he also discusses why (in his view), they are wrong.
 Conceptual Atomism and the Computational Theory of Mind

exist. But in such a world, you could still have a thought t* that was causally just
like t. The differences in content – the differences in respect of what kind of
thoughts they were – between t and t* would thus be without causal or, therefore,
psychological consequences. (After all, psychology is an empirical discipline, and
psychological categories – like all categories employed in natural science – are
causal.) There would be no psychological differences between t and t*, even
though they had different contents.
So, to close the argument, in some cases, a thing’s being a thought of a certain
kind is psychologically pregnant, and in other cases it is not. In virtue of being a
thought that 13 is prime, a brain-state falls into a psychological category that it
wouldn’t otherwise fall into. But it is not the case that, in virtue of being a thought
that Alpha Centauri is lovely, a brain-state falls into a certain psychological cate-
gory that it wouldn’t otherwise fall into. Thus, the property of being a thought
sometimes is, and sometimes is not, a psychological property.

This view is obviously ad hoc. And there is no need to take it, since the data can be ac-
commodated without it. We’ve already seen the reasons for this; but a brief review of
those reasons may be appropriate.
If one grasps Alpha Centauri, one does so by grasping some existence-claim that is
available to be grasped in any world w, regardless of whether Alpha Centauri exists in w
or not. In general, given any external object x, if one grasps x, one does so by way of some
existential proposition that can be grasped in any world, regardless of whether x exists
there or not. So we must consider the grasping of propositions of this kind – these pure-
ly qualitative, existential propositions – when we wish to come to a conclusion as to what
psychological properties a mental state has in virtue of having a certain content.
When we look at propositions of this kind, we discover the following important
fact. In virtue of being a grasping of such a proposition, a mental state (or brain-state)
does fall into this, as opposed that, psychological category. In other words, when we
look at these propositions, we see that the property of being a thought of a certain kind
is a psychological property – a property such that psychological explanation and clas-
sification are to be made in terms of it.
The content-externalist is forced to say that, even if P and P* are different proposi-
tions, the psychological properties one has in virtue of grasping the one may be indis-
tinguishable from the psychological properties that one has in virtue of grasping the
other. (To evade this, content-externalists have to take the tortured measure of saying
that a correct psychological classification of a person’s mental states may involve
knowledge of remote recesses of the cosmos.)
The content-externalist is also forced to make the following, equally counterintui-
tive claim. Supposing that two different people – say, Smith and Jones – both grasp P,
the psychological properties that Jones has in virtue of grasping P may be different
from the psychological properties that Smith has in virtue of grasping P.
A short story will help make it clear why content-externalism is committed to this
last claim. Apart from the differences about to be described, Smith and Jones are qual-
 Chapter 8.  The problem of de re senses 

itatively identical. In the morning, Smith sees Venus and, in so doing, he acquires a
concept of Venus. Of course, what Smith sees is a brightly lit object occupying a certain
place in the morning sky. He forms the belief that the thing he is seeing is lovely. So,
according to the externalist, there is some x, such that Smith now believes the singular
proposition: x is lovely.
In the evening, Jones sees Venus and, in so doing, he acquires a concept of Venus.
Of course, what Jones sees is a dimly lit object occupying a certain place, different from
the place just mentioned, in the evening sky. He forms the belief that the thing he is
seeing is lovely. So, according to the externalist, there is some x, such that both Smith
and Jones now believe the singular proposition: x is lovely.
But it is obvious that Smith’s belief will be psychologically and therefore causally
very different from Jones’ belief. Smith’s belief will lead him to believe that some object
in the morning sky is lovely. But ceteris paribus the same is not true of Jones’ belief.
Smith’s belief will lead him to form an intention to take a picture of the sky in the
morning. But ceteris paribus the same is not true of Jones’ belief.
So if content-externalism is right, there are some propositions P such that the
psychological properties that one person has in virtue of grasping P may be entirely
different from the psychological properties that some other person has in virtue of
grasping P.
But this is incoherent, since it is inconsistent with the well-known fact that, out-
side of the sub-atomic realm, qualitatively identical initial conditions lead to qualita-
tively identical outcomes – or, equivalently, that two initial conditions have qualita-
tively similar outcomes in so far as those conditions are themselves qualitatively
similar. (Of course, this may only be approximately true. But the differences between
Smith and Jones fall well outside the margin of approximation.)

Content-externalist counter-responses

Admittedly, the argument just given is less than probative; for the content-externalist
has a compelling reply to it:
There is some x, such that x is identical with Venus and such that, on the occasions
described, Smith and Jones both form the singular belief x is lovely. But that is not
the end of the story. When Smith sees Venus, he not only forms the belief x is
lovely; he also forms various other beliefs, e.g. there is an object in a certain part of
the morning sky having such and such color…And when Jones sees Venus, he not
only forms the belief x is lovely, but also various other beliefs, e.g. there is an object
in a certain part of the evening sky having such and such color…, where the color and
the location just mentioned are different from the ones mentioned in the descrip-
tion of Smith’s belief. So contrary to what you said, the content-externalist is not in
the position of having to explain the differences between Smith and Jones in terms
 Conceptual Atomism and the Computational Theory of Mind

of their both coming to believe x is lovely. In coming to believe that singular propo-
sition, they inevitably come to believe many other propositions, and the psycho-
logical differences between Smith and Jones are easily explained in terms of those
other propositions.

First of all, in this passage, the content-externalist is really saying that nothing is to be
explained in terms of the fact that Smith and Jones both grasp a certain singular proposi-
tion. By the objector’s own admission, that fact is completely inert in the way of account-
ing for differences (and also similarities) in their psychological and bodily properties. All
of those differences are to be explained in terms of the differences in the existential-de-
scriptive propositions that they grasp. (They are to be explained in terms of the fact that,
whereas Jones grasped some proposition involving the concept evening star, Smith
grasped some proposition involving the concept morning star.) So, if the content-exter-
nalist is right, the supposed fact that both Smith and Brown grasped x is lovely is without
any consequences as regards their psychologies. This is tantamount to saying that noth-
ing about those two people is to be understood in terms of the fact that they both grasped
that proposition. And that, in turn, is the same as saying they both don’t grasp it.
Conceivably, a content-externalist might counter-reply as follows:
Consider the descriptive propositions in terms of which the aforementioned differ-
ences between Smith and Jones are to be explained. Those descriptive propositions
are themselves object-involving. For example, consider the proposition: there is a
lovely object in that part of the sky…That proposition is object-involving with re-
spect to a certain part of a certain sky. So you haven’t yet shown that a grasp of ob-
ject-involving propositions is causally and therefore psychologically meaningless.

Yes, but an exact analogue of the argument that we just ran with respect to x is lovely will
apply to any singular proposition that is object-involving with respect to any external
object. Consider the proposition: there is a lovely object in that part of the sky…Given a
small amount of literary creativity, it is easy to hypothesize a scenario that demands that
we say about that proposition exactly what we said about x is lovely. And in that second
scenario, in order to explain the psychological differences between Smith and Brown,
we will have to cite the fact that, on the relevant occasions, the sky was given to Brown
in one way (i.e. via one descriptive content), whereas it was given to Brown in a different
way (i.e. via a different descriptive content). So, once again, when it comes time to ex-
plain the psychological differences between Smith and Brown, it is entirely irrelevant
what singular propositions they come to believe concerning the external world. Those
differences will have to be explained in terms of the fact that, on the relevant occasions,
the one person did, whereas the other did not, come to believe in the truth of some
proposition that was not object-involving with respect to anything external.
 Chapter 8.  The problem of de re senses 

The notion of a de re sense

Given these points, we are in a position to evaluate some of Evans’ most important
semantic and epistemological claims. Because these must be understood in terms of
Frege’s (1892) semantic system, let us briefly review the relevant aspects of that system,
along with our findings concerning it.
“The inventor of bifocals is identical with the first post-master general” expresses
something non-trivial, whereas “the inventor of bifocals is identical with the inventor
of bifocals” does not. Frege plausibly explained this by saying that, although they have
the same referent, “the inventor of bifocals” and “the first postmaster general” differ in
sense, a corollary being that a sentence containing the one defenite description will
ceteris paribus differ in cognitive value from a sentence containing the other.
Frege held that the sense-reference distinction also applied to proper names. The
sentence “Hesperus is Phosphorous” expresses something non-trivial, whereas the
sentence “Hesperus is Hesperus” does not. Frege concluded that proper names, like (in
his view) definite descriptions, have both sense and reference.
Kripke showed that Frege’s position is wrong as far as proper names are concerned.
(We’ve also seen that, despite its immense prima facie plausibility, Frege’s analysis of def-
inite descriptions involves an incoherent conflation of a number of different concepts –
that it fails to distinguish between semantics and pre-semantics and also between refer-
ence and quantification.) Like us, Evans believes that Kripke’s arguments are cogent, and
he thus believes that Frege’s analysis of “Socrates” and “Hesperus” is false. But Evans
(1985: Chapter 7) believes that we could invent proper names with respect to which
Frege’s analysis was correct. Evans holds that we could invent names – actual names, not
quantifiers disguised as names – that had Fregean senses for at least part of their seman-
tic contents. In connection with this, Evans’ argues there exist what McDowell would
later refer to as “de re senses.”
On the basis of these semantic contentions, Evans’ puts forth a far-reaching and, if
true, extremely important epistemological thesis, namely: there are de re modes of
presentation. This thesis is a fusion of three different bodies of thought: first, Frege’s
semantics and epistemology; second, Kripke’s work on proper names; and, third, con-
tent-externalism.
Here is what I would now like to show. First, contrary to what Evans says, there
cannot, even in principle, be proper names that have senses. The Fregean concept of
sense is an incoherent amalgamation of semantics and epistemology. Evans’ belief that
there are de re senses involves a similarly incoherent fusion of semantics and episte-
mology. But Evans’ view, unlike Frege’s, not only incoherently conflates semantics with
epistemology: it conflates semantics with an epistemology that is itself incoherent.
(Frege’s epistemology was not unlike that advocated in this work. Frege held that it is
not objects per se that are the constituents of the propositions that one grasps, but only
senses or concepts of objects. That view, we have argued, is entirely correct. Frege’s
 Conceptual Atomism and the Computational Theory of Mind

projection of this view into his semantics – viz. names have senses, not objects, for
their semantic contents – turned out to be false and, indeed, incoherent.)

Evans’ argument for the existence of de re senses6

Let us stipulate that “Julius” is to be defined as follows:

(E1) If there is a unique thing x such that x invented the zipper, then “Julius” refers to
x; and if nothing uniquely invented the zipper then “Julius” doesn’t refer at all.
So, by our stipulation, “Julius” is a referring term. At the same time, it obviously has a
sense, this being the concept zipper-inventor (or perhaps unique zipper-inventor). On
this basis, Evans arrives at two conclusions.
First, even though Frege was wrong with respect to actual proper names, the
sense-reference distinction applies to invented proper names, such as “Julius.”7 Second,
there are de re senses.
The idea behind the second claim is as follows. If somebody x uniquely invented
the zipper, then “Julius was tall” is true exactly if x invented the zipper. In other words,
given that, in actuality, x uniquely invented the zipper, the proposition encoded in that
sentence is true in a world w exactly if x was tall in w. At the same time, as we saw in
the last paragraph, “Julius” has a Fregean sense (unique zipper-inventor), and this
Fregean sense is part of the proposition encoded in that sentence.
So, Evans concludes, if x uniquely invented the zipper, “Julius was tall” encodes a
proposition that is object-involving with respect to x. At the same time, the reason why
that proposition has x as a constituent is that it contains a concept that is object-involv-
ing with respect to x. For any predicate phi, an exactly similar line of thought shows
that, if x uniquely invented the zipper, FJulius has phiG encodes a proposition that has x
as a constituent, the reason being that it comprises some concept that is object-involv-
ing with respect to x.
There is another fact about Evans’ proposal that must be pointed out. According to
Evans, the proposition encoded in “Julius was tall” will vary from world to world. If
Mary uniquely invented the zipper in w, then in w that sentence encodes a proposition

6. This argument is to be found in Evans (1982: 50-51, 1985: Chapter 10.) An excellent clarifi-
cation of Evans’ views is to be found in Grush (1999). But for Grush’s commentary, my level of
understanding of Evans’ work, especially in relation to the concept of a de re sense, would have
been much more tenuous than it is. See the chapter of Grush (1999) entitled “Guide to Chapter
6 of Gareth Evans’ Varieties of Reference.”
7. Evans (1982: 50-52) also holds that, on rare occasions, descriptive proper names come into
existence more or less naturally, putative examples being “Deep throat” and “Jack the Ripper.” So the
semantics for “Jack the Ripper” would be given by the rule: if somebody x uniquely committed the
White Chapel murders, then “Jack the Ripper” refers to x; and if nobody satisfies that condition, then
“Jack the Ripper” doesn’t refer at all and FJack the Ripper has phi G fails to encode any proposition.
 Chapter 8.  The problem of de re senses 

that is true iff Mary invented the zipper. For this reason, says Evans, the kind of sense
that “Julius” has isn’t quite the same as the senses of which Frege spoke. For Frege, the
proposition meant by “the inventor of bifocals snored” doesn’t depend on who in-
vented bifocals. This is because, in Frege’s view, that person is not himself a constituent
of the proposition meant by that sentence: only the sense of “the inventor of bifocals”
makes it into that proposition. Matters are different with “Julius.” Even though that
expression has a sense, the proposition meant by “Julius was tall” does depend on who
invented the zipper in the world of utterance.
The semantics of “Julius” can be understood in terms of Kaplan’s (1975) dthat-op-
erator. If x is the unique phi, then by definition, Fdthat [the phi] G refers directly to x; and
the proposition meant by Fdthat [the phi] has psiG is true exactly if x has psi. So suppos-
ing that Franklin invented bifocals, the proposition encoded in “dthat [the inventor of
bifocals] snored” is true in any world W where Franklin snored, it being irrelevant
whether Franklin, or anyone else, uniquely invented bifocals in W. “Julius” has the
same semantics as “dthat [the person who invented the zipper].”
Fregean senses fail to incorporate the things that they pick out into the propositions
of which those senses are a part. The sense makes it into the proposition, but not the
referent. But the sense of “Julius” does manage to incorporate the referent of “Julius” – the
actual person – into the proposition. So Evans-style senses don’t merely describe, al-
though that is obviously part of what they do. They also absorb the relevant objects into
themselves. That is why – depending on whether it is Mary, Fred, or Ethel who uniquely
invented the zipper – the sense of “Julius” manages to make Mary herself or Fred himself
or Ethel herself be a constituent of the proposition meant by “Julius was tall.” In general,
Evans-style senses actually comprise the objects that they describe. “Julius” thus has a de
re sense: a sense that actually comprises the object that it picks out.
It follows that what the sense of an utterance “Julius” is depends on who invented the
zipper in the world in which that utterance occurs. If Mary invented it, then “Julius” will
have one sense. If Fred invented it, that expression will have a different sense. Those
senses will be similar, but not identical. More exactly, they will be exactly alike except that,
where the one has Fred as a constituent, the other has Mary as a constituent. In general,
two senses might be exactly the same except that, in the place (so to speak) where one of
them had external object x as a constituent, the other had external object y.
If cogent, the line of thought just discussed has two important consequences. First,
even though the sense-reference distinction is not in fact true of actual proper names,
it is true of possible proper names. Second, some senses (e.g. the senses of expressions
like “Julius”) are de re with respect to external objects.
If correct, these two points would at least partially reconcile Frege’s semantics with
the contemporary semantic view that, in at least some cases, the propositions meant by
sentences have actual bits of the external world for their meanings. So supposing that
Evans’ views are correct, we can hold onto some powerfully argued contemporary views
about semantics, without foregoing Frege’s extremely plausible distinction between sense
and reference.
 Conceptual Atomism and the Computational Theory of Mind

The epistemological underpinnings of the notion of a de re sense

The view that there are de re senses is the semantic counterpart of an epistemological
thesis. We’ve already discussed this thesis; but a brief summary of our findings may be
appropriate. In the morning, you look at what is in fact Venus. And in the evening, you
(again) look at what is in fact Venus. But, of course, you don’t necessarily realize that
the thing you saw in the morning is the same as the thing you saw in the evening. This
is obviously because your morning-perceptions make you aware of Venus through one
mode of presentation (one that, if verbalized, would be along the lines of: “last celestial
body to disappear from the morning sky”), whereas your evening perceptions make
you aware of Venus through a different mode of presentation (one that, if verbalized,
would be along the lines of: “first celestial body to appear in the evening sky”). It is
because these modes of presentation differ that you can coherently question whether
you were seeing the same thing on both occasions.
At the same time, content-externalism tells us that, in both cases, Venus itself is a
constituent of the contents of your perceptions. Evidently, then, the modes of presenta-
tion implicated in your perceptions not only presented Venus to you, but managed to
assimilate Venus into themselves. Those modes of presentation – those descriptive
contents – were therefore de re with respect to Venus.

Evaluating Evans’ argument

The notion of a de re sense is incoherent, as is the notion of a de re mode of presentation.

Suppose that, in actuality, x uniquely invented the zipper. Let P be the proposition
that, according to Evans, is encoded in “Julius was tall.” (Here we needn’t distinguish
between types and tokens.) According to Evans, P is true in a world w exactly if, in w,
x is tall. So P is identical, or at least equivalent, with x is tall. Obviously the concept
unique zipper-inventor is no part of x is tall or, with a few irrelevant exceptions, of any-
thing logically equivalent with x is tall.
Before proceeding, let me explain what these “irrelevant exceptions” are. Obviously
x is tall is logically equivalent with either x is tall or a unique zipper-inventor was a square-
circle. But that proposition clearly isn’t what is meant by “Julius was tall”; and, for obvi-
ous reasons, the same is true of any other proposition that is both logically equivalent
with x is tall and also has the concept unique zipper-inventor has a component.
So, despite Evans’ own stipulation, the proposition meant by “Julius was tall” does
not have the concept unique zipper-inventor as a component. That sense is nowhere to
be found in that proposition.
What are we to say about the idea that P has as a component some de re mode of
presentation – about the idea that one of P’s components is a mode of presentation that
actually has x as a constituent?
 Chapter 8.  The problem of de re senses 

That idea is shown to be untenable by an argument similar to the one just given.
Supposing that such a mode of presentation exists, there is no denying that at least part
of the content of that mode of presentation is the concept unique zipper-inventor. But
we’ve just seen that, with a few irrelevant exceptions, no proposition that has the same
truth-conditions as P has that concept as a component. Thus, the concept unique zip-
per-inventor isn’t a component of any proposition that has the same truth-conditions
as the proposition that by Evans’ own stipulation is the meaning of “Julius was tall.”
Evans has thus failed to identify any proposition that has a de re sense as a component.
He has not shown that there are de re senses.
It is not hard to see what led Evans astray. Somebody who hears an utterance of
“Julius snored” or “Julius was wealthy” must work through the semantic rule that as-
signs a referent to “Julius.” Supposing that Brown is the unique zipper-inventor, that
rule is not: “Julius” refers to Brown. Rather it is: for anyone x who uniquely invented the
zipper, “Julius” refers to x. For reasons given earlier, this pre-semantic fact easily ex-
plains why “Julius snored” conveys the proposition some unique zipper-inventor snored.
But, by Evans’ own stipulation, the proposition literally meant by “Julius snored” is
true in worlds where nobody invented the zipper. Consequently, that proposition has
nothing to do with zipper-inventing. What does have to do with that concept is the
proposition through which one computes the meaning of “Julius snored.” Evans is
committing the now familiar error of conflating semantics with pre-semantics.8

8. Here one might make the following reply on Evans’ behalf:

“For a moment let us forget about the niceties of Evans’ own argument. Suppose we stipu-
late that ‘Julius’ is a device of singular reference and that ‘Julius’ has the concept unique zipper-
inventor as a component. Given this, let P* be the proposition semantically encoded in ‘Julius
was tall.’ P* would have x himself as a constituent; and P would also have the concept unique-
zipper inventor (and possibly also some concept built out of that concept) as a component. In
this case, it seems to me, P* might well be the kind of proposition that Evans had in mind: a
proposition that comprises a sense that is de re with respect to some external object.”
But in that case, the question is: which proposition is P*? P* cannot be x was tall. After all,
by hypothesis, P* must be a proposition which has the concept unique zipper-inventor as a com-
ponent. But P* cannot be identical, or logically equivalent, with:
(#) something y was a unique zipper-inventor and y was tall.
For neither (#), nor any proposition logically equivalent therewith (with a few irrelevant excep-
tions, to be identified in a moment), has x as a constituent. Further, (#) doesn’t have the same
truth-conditions as the proposition that, by Evans’ own stipulation, is meant by “Julius was tall.”
Also, if (#) were the meaning of “Julius was tall”, then that sentence would have a quantified ge-
neralization (or, at least, a proposition logically equivalent with such a generalization) for its
meaning. And in that case, as we saw earlier, “Julius” would be a quantifier, contradicting Evans’
stipulation that it is a device of reference.
Of course, there are propositions that are both object-involving with respect to x and have
the concept unique zipper-inventor as a constituent, for example:
(^) x was tall and something y was a unique zipper-inventor and y was tall.
 Conceptual Atomism and the Computational Theory of Mind

De re modes of presentation and Kripke’s paradox

In the morning, you look at what is in fact Venus. You sincerely say “that thing is
lovely.” Let P1 be the proposition your knowledge of which motivated this utterance.
That evening, you again look at what, unbeknownst to you, is in fact Venus. You sin-
cerely say “that thing is not lovely.” Let P2 be the proposition your knowledge of which
motivated this utterance.
It is a datum that, given only what we’ve said, you are not necessarily irrational. The
reason is obviously that the object in question was given to you in different ways – ways
that are so different that you don’t know that you are describing some one object as both
“lovely” and “not lovely.”
But the content-externalist says that, for some x, you are thinking x is lovely in the
morning and x is not lovely in the evening. Astonishingly, what content-externalists
have done in response is to revise our conception of rationality: they hold that one can
assent to a proposition and its negation without being irrational. In their view, I can
believe P and not-P without being irrational.9
Given what we have seen, there is no need to embrace this implausible, if not tau-
tologically false, view. The content of your morning-perception is an existence-claim
that is satisfied by Venus; the same is true of the content of your evening-perception.
But Venus is not itself a constituent of the content of either perception, and there is no
x such that you believe the singular proposition x is lovely and also the singular propo-
sition x is not lovely. The semantic externalist is quite right to say that you have affirmed
both of those propositions. But we’ve already seen why this doesn’t entail that you be-

But (^) doesn’t have any de re sense as a constituent. Further, to echo a point made a moment ago, (^)
isn’t logically equivalent with the proposition that, according to Evans’ own stipulation, is meant by
“Julius was tall.” Most importantly, it is pretty clear that (^) simply isn’t what Evans had in mind. (So
(^) is one of the “irrelevant exceptions” mentioned a few lines back.)
Also, if (^) were the meaning of “Julius was tall”, then “Julius” would be both a singular term
– one that picks out x – and something that has a sense (unique zipper-inventor). But it would
not have both of these properties by virtue of having a de re sense. It would have both those
properties by being a device of singular reference and a quantifier. But these two roles wouldn’t
fuse together into a de re sense. “Julius” would simply be an orthographically condensed repre-
sentation of two entirely distinct semantic functions.
Of course, Evans didn’t think that (^) in particular was the meaning of “Julius was tall.” Nor
did he believe that, if (^) were the meaning of that sentence, “Julius” would be of any particular
theoretical interest. But what we said about (^) is true of any proposition that has both x and the
concept unique zipper-inventor as constituents. Any such proposition will lack a de re sense. And
if any such proposition is the meaning of “Julius was tall”, the result is not that “Julius” is a sense-
bearing singular term – the result is not some kind of fusion of description with reference. Ra-
ther, the result is that “Julius” is a phonetically condensed way of encoding two quite separate
semantic functions – reference and, separately from that, description – and that consequently
fails to illustrate anything non-trivial concerning actual or possible languages.
9. See Kripke (1979), Salmon (1986), Neri-Castañeda (1989), Kaplan (1989).
 Chapter 8.  The problem of de re senses 

lieve either of those propositions. What you believe are existence-claims that describe
those singular propositions, not those singular propositions themselves. Since those
existence-claims are perfectly consistent with one another, we don’t need to jettison
the truism that it is irrational to believe P and not-P.

Refining our analysis

There is an apparent problem with the view just presented. According to our analysis,
our awareness of the external world is completely descriptive. If I am aware of a given
rock, it is because I know an existence-claim that describes it. I cannot possibly be ac-
quainted, in the strict Russellian sense, with the rock.
But, some will say, knowledge by description presupposes knowledge by acquaint-
ance. After all, one must grasp the components of the descriptions that one grasps. (I
couldn’t grasp the description inventor of the fugue unless I grasped the concepts in-
ventor, fugue and so on.) And, arguably, it would be viciously regressive to suppose that
those components were always grasped by description. But our analysis would appear
to be guilty of exactly this sort of regressiveness, given that it doesn’t allow us to be
acquainted with anything, or at least not with anything external.
One possible counterargument would be to deny that knowledge by description
does in fact presuppose knowledge by acquaintance. This, I believe, is the position
anti-foundationalists such as Sellars and Bonjour would take. But it behooves us to
produce a counterargument that doesn’t assume the truth of anti-foundationalism,
given how controversial that doctrine is. Just such a counterargument is to be found in
an incisive analysis put forth by John Searle (1983: 225-230). I have modified Searle’s
analysis so as to integrate it into our own.
Suppose that you see Venus in the morning. Let V be the visual sensation mediat-
ing this perception. There are two points to make about V. First, you are actually ac-
quainted with V. (In any case, this is what Russell thought, and we have found many
reasons to accept Russell’s view.) Second, even content-externalists will grant that V is
partly descriptive – they will grant that it gives you existential information of the kind
that we have so often described. (Content-externalists will also hold that V gives you
non-descriptive, object-involving information. Therein lies the disagreement between
externalists and internalists.)
In light of this, consider the existence-claim:
(*) There is some external state of affairs x such that x matches the description
coded in V, and such that x caused V.
If V is caused by a blow to the head, or by consumption of a psychedelic drug, then it won’t
be a perception of anything. The cause of V must be something which fits the description
coded in V. It must be a brightly lit sky comprising a lovely luminescent planet.
 Conceptual Atomism and the Computational Theory of Mind

But even this, though necessary for V’s being a perception (as opposed to a coin-
cidentally correct hallucination), is not sufficient. In other words, for V to be a percep-
tion, it is not enough that V be caused by something x that fits the description coded
in it. Suppose that a brightly lit sky causes Bob to punch you, which in turn causes you
to have V. In that case, V wouldn’t be a perception, even though it was caused, albeit
indirectly, by something which fits its content. So there is some specific mode of causa-
tion C such that, for V to be a perception, it is necessary and sufficient that V be caused
in manner C by something x that fits the description coded in it. (It is a question
greatly debated among epistemologists what C is. I will not attempt to answer this
question. But it may reasonably be taken for granted that there is some such privileged
causal relation.)
Given this, let us replace (*) with:
(**) There is some external state of affairs x such that x fits the description coded
in V, and such that x caused V in manner C.
Suppose that SA is the state of affairs which uniquely satisfies (**). We just saw that, in
that case, V is a perception of (**) SA. Of course, SA will comprise Venus, and it will also
include various parts of the sky, including S. It will also, we may suppose, include various
birds, trees, and so forth.
So V will be a perception of Venus, of S, and of those various birds and rocks. At
the same time, none of those things will be a constituent of V’s content. They will be
described by it, but not constitutive of it. The fact that you are seeing that particular
bird, as opposed to its doppelganger, is indeed a function of the fact that you are caus-
ally connected to that bird in a way in which you are not causally connected to its dop-
pelganger. But the connection is still a relation of description. That bird is, whereas its
doppelganger is not, a part of the unique state of affairs which satisfies (**). (As before,
let SA be that state of affairs.) Given what (**) is, in order to satisfy that existence-
claim, SA must cause V (and must do so in a quite specific way). At the same time, it is
clear that (**) describes SA, and doesn’t have SA as a constituent. The reason you are
seeing that particular bird and that particular planet, and not their doppelgangers, is
that they are parts of SA, that being the thing which satisfies (**). So the reason that
you see that particular bird is that there is some description D such that D is uniquely
satisfied by that bird and such that you know that D is uniquely satisfied. (The proposi-
tion that you know is not that bird satisfies D. After all, a precondition for grasping that
proposition is that you have a concept of the bird in question. The relevant proposition
is simply: D is uniquely satisfied. D is a description that is, in fact, uniquely satisfied by
the bird in question; and it is by way of your knowing that D is uniquely satisfied that
you grasp that bird.) At the same time, your knowing that description involves your
having a quite specific causal relation to that bird.
 Chapter 8.  The problem of de re senses 

Some virtues of this analysis

The model just proposed satisfies all the relevant desiderata. It is consistent with the fact
that our awareness of the external world is descriptive. For reasons that we have discussed,
it follows that our analysis, unlike content-externalism, is consistent with the causal potency
of the mental. Further, our analysis is consistent with the fact that conception involves a
causal connection between subject and object and also with the reasonable, though debat-
able, presumption that knowledge by description presupposes knowledge by acquaintance.
Our analysis has one other virtue, which the following story will help bring to
light. Max and Twin-Max are atom-for-atom duplicates of each other that live in the
same world. If Max were to describe everything that he believed, felt, and so forth, and
Twin-Max were to the do the same, the resulting descriptions would be completely
indistinguishable. Thus, so far as mental content is strictly descriptive, Max’s mental
content cannot distinguish him, either qualitatively or numerically, from Twin-Max. It
would therefore seem that Max’s thoughts couldn’t possibly have different truth-mak-
ers from Twin-Max’s, given that the truth-maker of a thought is some state of affairs
that is described by that thought.
But it is a plain fact that Max’s thoughts do not, at least not categorically, have the
same truth-makers as Twin-Max’s. Suppose that, at time t, Max and Twin-Max both
have intense headaches – let x and y be these headaches – and that each is thinking that
his headache is intense. Max’s thought is made true by x’s being intense, whereas Twin-
Max’s corresponding thought is made true by y’s being intense. But it isn’t just when
Max and Twin-Max are thinking about their own sensations that the one person’s
thoughts have different truth-makers from the other person’s corresponding thoughts.
For Max’s sense-perceptions and subsequent beliefs might have different truth-makers
from Twin-Max’s corresponding sense-perceptions and beliefs. If Max’s visual percep-
tion and subsequent belief are made true by Mary’s sitting on nearby sofa S, Twin-
Max’s corresponding perceptions and beliefs may be made true by some other person’s
sitting on some other sofa.
We’ve argued that our awarenesses of external objects are entirely descriptive: one
has thoughts about external object X by grasping some description that singles X out.
But we’ve just seen that no purely descriptive awareness could distinguish a given ob-
ject from some qualitatively identical but numerically distinct object. (Strawson (1959)
and Kaplan (1989) make precisely this point.) At the same time, we don’t want to give
up on the idea that our awarenesses of external objects are descriptive, since we’ve seen
that the opposing view has consequences too absurd to accept.
Our application of Searle’s theory solves this problem. One’s awareness of the ex-
ternal world is fundamentally descriptive. But the same does not hold with respect to
one’s awareness of one’s own (conscious) mental states. One can be acquainted with
such states. (If you burn your hand, you are not aware of the subsequent pain by know-
ing some existence-claim that describes it: you are directly aware of it.) x is a veritable
constituent of Max’s thought that his headache is intense, and y is a veritable constituent
 Conceptual Atomism and the Computational Theory of Mind

of Twin-Max’s corresponding thought. That is why, otherwise indistinguishable though

they are, those thoughts have different truth-makers. V is a veritable constituent of the
content of the existence-claim encoded in your perception of Venus. That is why, oth-
erwise indistinguishable though they are, a perception that doesn’t have V as a con-
stituent could have a different truth-maker from your Venus-perception. A similar line
of thought explains why Max’s visual perceptions and subsequent beliefs have different
truth-makers from Twin-Max’s corresponding perceptions and beliefs.
Even though all awarenesses of external objects are strictly descriptive, the same is not
true of all awarenesses of internal objects. We have non-descriptive, direct awareness of
some of our own mental states. The descriptions through which we grasp external objects
are typically constructed, in part, out of non-descriptive awarenesses of our own mental
states. If an awareness of this kind is a constituent of one perception, but not of some other,
then those two perceptions will differ in content, even if they are otherwise indistinguisha-
ble and even if, consequently, they coincide in respect of how they describe the external
world. In this way, we avoid the indefensible view that we have non-descriptive awarenesses
of external objects, and we also manage to explain how descriptively identical awarenesses
can have different objects.

The Searle-McDowell debate

In light of these points, we can adjudicate a dispute between John McDowell (1998: 260-
274) and John Searle (1983: 225-230). As we just saw, Searle advocates an internalist
epistemology: he rejects the idea that external objects can be actual constituents of the
contents of our thoughts, and he also holds that concepts of external objects involve
knowledge of truths that have bits of one’s own consciousness as constituents.
McDowell’s rejects Searle’s position, and he does so on the following grounds (this
is a paraphrase, not a quotation):
Searle grants that one can be aware of propositions that are object-involving with
respect to one’s mental states – specifically with respect to the sensations that me-
diate one’s perceptions. So why not grant that we can be aware of propositions that
are object-involving with respect to external objects? It seems arbitrary to believe
the one thing but not the other. So Searle’s grounds for advocating internalism are,
in fact, grounds for accepting externalism.

It is obvious what the fallacy in this argument is. Contrary to what McDowell assumes,
it is not arbitrary to hold that one can be acquainted with some of one’s own mental
contents but not with external objects. Let x be some extremely unpleasant itch that
you are having at time t, and let y be some extremely pleasant tickle that you are having
at that same time. At t, you are (or at least could be) acquainted with x and y and you
can therefore grasp propositions that are object-involving with respect to them. Let P
 Chapter 8.  The problem of de re senses 

and P* be two such propositions. More specifically, let P be the proposition if I take my
shirt off, x will go away, and let P* be the proposition if I take my shirt off, y will go away.
Of course, P and P* are exactly the same except that, in the place where one of them
comprises x, the other comprises y. It goes without saying your grasping P at time t
would have very different effects from your grasping P* at that same time. Thus, the
supposition that you are actually acquainted with x and y is consistent with the sup-
position that two propositions are distinct only if (ceteris paribus) a grasp of the one is
causally different from a grasp of the other.
But the view that we are acquainted with external objects is not consistent with
that supposition. In and of itself, a grasp of the proposition Mary is sitting on the couch,
sipping tea is (ceteris paribus) causally identical with a grasp of the proposition Twin-
Mary is sitting on the couch, sipping tea. We saw this in Chapter 2. It is only if one grasps
those propositions through different bodies of descriptive information that a grasp of
the one is causally different from a grasp of the other. But this shows that it is only so
far as Mary and Twin-Mary are described to us that a grasp of a proposition that is
object-involving with respect to the one is causally different from a grasp of a proposi-
tion that is object-involving with respect to the other. In other words, the supposition
that we can be non-descriptively aware of Mary and Twin-Mary, i.e. that we can be
acquainted with them, is inconsistent with the fact that propositions are different only
if a grasp of the one is (ceteris paribus) causally different from a grasp of the other. So,
contrary to what McDowell says, it is in no way arbitrary to hold that we can be ac-
quainted with (some of) our own mental contents, but not with constituents of the
external world.
 chapter 9

Publicity problems and the nature

of linguistic communication

We’ve seen that understanding “Socrates”, or any other referring term, involves associ-
ating it with some description that its referent uniquely satisfies. But we’ve also seen
that the descriptions that a person associates with such an expression are specific to the
circumstances under which he first encounters it. A consequence is that any two people
will almost certainly associate different descriptions with (for example) “Socrates.” But
surely “Socrates” is useless as a device of communication unless people mean the same
thing by it, and surely people cannot mean the same thing by it if any two people associ-
ate different descriptions with it. So our analysis seems to conflict with the fact that
referring terms can be understood only in so far as the descriptions associated with
them are shared.
This is a point that Frege (1892) stressed: the “senses” of referring expressions
must be public. If the sense that I associate with “Socrates” is great hemlock-drinking
philosopher, and the sense that you associate with it is ironic main character in most of
the Platonic dialogues, then (so one would think) what you mean by “Socrates was
wise” will be different from what I mean by those same words. What you mean will be
equivalent with: somebody x was uniquely an ironic main character in most of the pla-
tonic dialogues; and x was wise. And what I mean will be equivalent with: somebody x
was uniquely a great hemlock-drinking philosopher; and x was wise. According to Frege,
to the extent that people associate different senses with a given expression, that expres-
sion fails as a way of allowing people to express their thoughts to one another.
At the same time, it is a datum that we can use expressions like “Socrates” to make
ourselves understood to one another. My analysis seems to be inconsistent with that
fact. I believe that Socrates was a person of principle. The occurrence of “Socrates” in
the last sentence expressed that belief. But if my analysis is right, how can that be? For
what I mean by “Socrates was a person of principle” is one existence-claim, and what
you mean by it, and thus take it to mean when hearing it from others, is some entirely
different existence-claim.
I would now like to show why my analysis is consistent with the facts concerning
the usefulness of “Socrates”, and linguistic expressions generally, as devices of com-
munication. Our analysis is consistent with the important points made by Frege con-
cerning publicity and communicability. Indeed, I will go further than this. I will argue
that what enables people to communicate with one another is precisely that there is a
gulf between literal and cognitive content. Theories that don’t posit such a gulf make
 Conceptual Atomism and the Computational Theory of Mind

linguistic communication impossible. If the literal meanings of your words collapse

into the contents of the thoughts that lie behind them, then I will never be able to un-
derstand what you are saying. This, at any rate, is what I will now argue.1

Speaker’s meaning versus linguistic meaning

Let T be a token of “Socrates was wise.” As we’ve seen, a given person must access the
literal meaning of the proposition literally meant by T through some proposition like
the following:
(TP) There is some occurrence x of “Socrates” such that, five minutes ago, while chat-
ting with me in this pub, Fred uttered x, and such that, for some y, x uniquely
refers to y, and (finally) such that “Socrates was wise” has for its meaning the
proposition: y was wise. .
Obviously (TP) doesn’t say anything that is of general interest. It doesn’t say anything
that historians or philosophers would find illuminating.
As we’ve seen, the thoughts through which you compute the meanings of utter-
ances are of the same basic kind as the thoughts which prompt you to produce such
utterances. Since there is no way for you to understand T except by way of your knowl-
edge of some egocentric, contextual proposition like TP, knowledge of such a proposi-
tion must be what prompts you to utter T.
But this doesn’t entail that what you mean when you utter “Socrates was wise” is
TP. For reasons that we’ve given, when you say “Socrates was wise”, you know that what
is literally meant by your words is confined to the underlined part. You thus know that,
so far as your objective is to produce an utterance whose literal meaning is correct,
your objective is met exactly if that operator-right proposition is correct. For you to
mean that proposition by your utterance of “Socrates was wise” is simply for you to
make that utterance with that objective in mind.
It is true that you don’t directly grasp that proposition: you grasp it only in the
indirect sense described earlier. But you can still mean that exact proposition by your
words; you can mean x was wise, for the appropriate x. You don’t have to mean some
existence-claim, even though, ultimately, you are incapable of grasping anything other
than such a claim.
A digression might help clarify this last point. If you point to an object and say “that
is Rover”, your objective is only to label some object. More formally, there is some object
O such that your intention is to label O as “Rover.” For reasons that we have discussed, you
cannot grasp O except by way of a perception that describes it, and you therefore cannot
grasp O except by way of your knowledge of some proposition of the form: something
uniquely has phi. But, quite obviously, you can still intend that whatever it is that uniquely

1. Much of what I will say here is anticipated by Russell (1917: 152-167).

 Chapter 9.  Publicity problems and the nature of linguistic communication 

satisfies that existence-claim be labeled as “Rover.” An exact analogue of this point applies
to your relationship to the proposition x was wise. You do not grasp that proposition ex-
cept by way of grasping some other proposition that describes it – some existence-claim
that it uniquely satisfies. (Let P be this other proposition.) But you can obviously believe
that whatever it is that uniquely satisfies P is true; further, you can obviously express that
belief; and, finally, you can obviously utter a sentence with the intention of expressing that
belief. If you utter a sentence with that intention then what you mean by your utterance is
the singular proposition x was wise. It is therefore perfectly easy to explain how we can
mean propositions that we are not acquainted with. We can mean such propositions be-
cause we are acquainted with propositions that describe them.
Of course, there will probably be some resistance to our view that, when someone
speaks, what he means may be some proposition that he isn’t even acquainted with. If I am
not mistaken, the idea underlying this resistance is similar to, if not identical with, the idea
that forms of the heart of what is probably the most famous analysis of linguistic meaning
– that of H.P. Grice (1957). And we need only point out the well known problems with
Grice’s theory to see why, initially plausible though it is, that idea is false and why, conse-
quently, there oughtn’t be any resistance to our position.
According to Grice, expression-meaning must be understood in terms of speaker’s
meaning. “Snow is white” means snow is white in virtue of the fact that people mean snow
is white by “snow is white.” In general, our utterances mean what we mean by them.
Grice’s view conflicts with our view that, in uttering a sentence, one can mean a
proposition with which one is not acquainted. An exposition of Searle’s (1969: 43-47)
devastating criticism of Grice’s view makes it clear why this is so. You cannot check-
mate somebody in a context where the rules constitutive of the game of chess are inop-
erative, and you cannot even intend to checkmate somebody unless you believe that
those rules are operative. Similarly, you cannot affirm that snow is white with the
words “snow is white” in a context where the rules constitutive of the English language
are inoperative, and you cannot even try to affirm that proposition with those words
unless you believe that those rules are operative. A speaker cannot successfully affirm
P with utterance S unless S already means P; and a speaker cannot even try to affirm P
with S unless he believes that S already means P.2 So the concept of speaker’s meaning

2. Of course, you could mean snow is white by “snow is white” even if you that thought that
the literal meaning of that sentence was the proposition that grass is green. A person’s boss can
mean you’re fired with the words “I can’t guarantee that your future here will be a good one”,
knowing full well that those words don’t literally mean you’re fired. But the boss knows that given
what they literally mean, those words will convey that message under those circumstances. So
the boss is able to mean you’re fired by those words because he knows that there is a pre-existing
symbolic relationship (albeit not one of literal meaning) between those words and that message.
Therefore, we don’t here have a meaningful exception to the principle that speaker-meaning is
to be understood in terms of expression-meaning. So far as we seem to have such an exception,
that is because we are conflating the concept of literal meaning, which is a very special kind of
meaning, with the concept of meaning tout court.
 Conceptual Atomism and the Computational Theory of Mind

must be understood in terms of that of linguistic meaning, not vice versa. Speaker’s
meaning presupposes conventional meaning, as Searle puts it.3
In Chapter 25, we will further discuss the problems with the Gricean view. But for
now let us proceed on the reasonable assumption that it is false. (See Searle 1969: 42–
50 and Blackburn 1984: 110–118 for cogent arguments against the Gricean view.)
But Grice was surely right to this extent: what people think – what is going on in
their minds – is obviously relevant to what comes to be meant by the noises and ink-
marks they produce.4 This must be understood aright. If I were to mean snow is white
with the sounds “grass is green”, the latter would not come to mean the former. But the
fact that “grass is green” means what it does is a reflection, albeit an indirect one, of
what was going on in the minds of some people at some point (or, more likely, various
points) in human history. From this last fact, however, it doesn’t follow that speaker’s
meaning – the phenomenon of a person’s meaning something by a sound or ink-mark
– is prior to semantic meaning. What follows is that speaker’s thinking is prior to se-
mantic meaning. The Gricean model starts with the correct point that what we think is
at least partially determinate of what our words mean; but it conflates this truth with
the falsehood that what we mean determines what our words mean.
A brief summary will help us close the argument. Grice holds that speaker’s mean-
ing must be understood in strictly psychological terms. We hold that, in uttering a
sentence, one can mean a proposition with which one is not acquainted. Even though
they may not technically be incompatible, these two views clearly conflict. But speaker’s
meaning is not to be understood in strictly psychological terms. It turns out that a vast
network of (putative, if not actual) social interactions mediates between the speaker’s
thoughts, on the one hand, and the speaker’s meaning, on the other. Given this last
point, there turns out to be nothing counterintuitive about our contention that, in ut-
tering a sentence, what one means may be a proposition with which one is not ac-
quainted. We’ve seen that one can have knowledge by description of propositions with
which one is not acquainted. It is not out of the question that, where some such propo-
sitions are concerned, the relevant descriptions have a metalinguistic component and,
more exactly, have the form the proposition literally meant by sentence S. Supposing this
to be the case, one’s intention, in uttering sentence S, could be to affirm the proposition

For exactly similar reasons, one could mean snow is white by an utterance of “snow is white”
without believing that the latter literally meant the former. But this would be possible only if the
speaker believed there to be some kind of pre-existing symbolic relationship between those sounds
and that proposition. So we don’t have a genuine exception to the principle that speaker’s meaning
is to be understood in terms of expression-meaning; and, just as we asserted, the concept of spea-
ker’s meaning is to be understood in terms of the concept of literal meaning, and not vice versa.
3. Wittgenstein (1958 § 510) observed that you cannot say “snow is white”, but mean grass is
green. This was meant to corroborate his view that linguistic meaning is to be understood in (at
least partially) non-psychological terms. This argument of Wittgenstein’s appears to be cogent.
4. Wittgenstein (1958) seems to deny this. We will consider his views in Chapter 17.
 Chapter 9.  Publicity problems and the nature of linguistic communication 

literally meant by S, whatever that proposition should be. In other words, one could have
knowledge by description (but not knowledge by acquaintance) of the proposition
meant by S; further, one could also believe that proposition, whatsoever it is, to be cor-
rect; finally, one could, under those circumstances, utter S with the intention of impart-
ing that belief. It seems to me that, if one uttered S with such an intention, one would,
in so doing, mean what one was saying, even though one was not acquainted with the
proposition meant.
To sum up, once it is seen that, contrary to what Grice held, the phenomenon of
speaker’s meaning is not to be understood in strictly psychological terms, and that it is
to be understood in terms of one’s relation to social institutions that have only a partial
(and quite limited) basis in one’s psychological states, there is no longer any reason to
reject our analysis of speaker’s meaning. In any case, there is no longer any reason to
reject it on the grounds that, if it is correct, what one means in uttering a sentence may
sometimes be a proposition of which one has only descriptive knowledge.
Let us now discuss the epistemological basis of these points. The expressions of a
language are public entities. Their occurrences are given to us through sense-perception,
and the same is true of the activities which give them their meanings. How do you learn
what “Hesperus” refers to? Somebody points to an object and says “that is Hesperus”, or you
overhear somebody using that expression. In any case, it is always ‘ultimately’ through
sense-perception that you learn what that term refers to.
As we’ve discussed, for you to have a sense-perception of an object is for that ob-
ject to be described to you. Given this point, along with the fact that it is through sense-
perception that you learn semantic rules, it follows that those semantic rules are always
described to you.
Of course, the perceptually encoded description through which a semantic rule is
given to you will incorporate the idiosyncrasies of your particular situation, and will
not have any basis in the semantic rule itself. Those idiosyncrasies are reflected in the
information that is conveyed to you by utterances of sentences like “Socrates was wise”
and “Hesperus is lovely.” As a result, there is categorically a lack of fit between the se-
mantic rules themselves, on the one hand, and the contents of our thoughts, on the
other. It is thus of the essence of any public language that what its expressions mean not
fit too snugly with the thoughts that we have.
This does not mean that linguistic communication is not possible. It is a datum
that people successfully transmit information through language, and that there is such
a thing as using language to make oneself understood. What the preceding line of
thought shows is that we must reanalyze what it is for someone to successfully com-
municate something. Let us now supply the needed re-analysis.
 Conceptual Atomism and the Computational Theory of Mind

The structure of linguistic communication

Suppose that Smith and Jones are both cognitively normal competent speakers of English.
Of course, both know what is meant by “Socrates is wise.” But, as we’ve seen, what Smith
knows may be different from what Jones knows. What Smith knows has the form:
(1) there is some x such that x has properties P1…Pn and “Socrates was wise” is
true exactly if x was wise.
And what Fred knows has the form:
(2) there is some x such that x has properties P*1…P*n and “Socrates was wise” is
true exactly if x was wise,
where P1…Pn may be, and probably are, entirely different from P*1…P*n.
But (1) and (2) have the same truth-maker. The thing which satisfies (1) is identi-
cal with the thing which satisfies (2). So even though what Fred takes away from Smith’s
utterance of “Socrates was wise” may be very different from the thought that prompted
Smith’s utterance, there is still a significant sense in which a truth has been transmitted.
Thanks to that utterance, Fred now has a thought that has the same truth-maker as
Smith’s thought.
Another illustration may be in order. I see Brown for the first time at time t. But
what I see is not just Brown; what is given to me in my perception is some existence-
claim along the lines of: over there, next to that tree, there is some man x such that x is
unshaven and looks disheveled; moreover, x just introduced himself to me by saying “I am
Brown: please forgive my appearance – you see, I didn’t sleep well last night…”
Let us continue this story. You see Brown for the first time at time t*, and the con-
tent of your perception is given to you through the existence-claim: behind that podi-
um, there is a clean-shaven, debonair fellow x such that x is giving a competent, but ulti-
mately not very informative, lecture on the history of psychology; moreover, I’ve been told
that x’s name is “Brown.”
Suppose that, at some later time, you and I are at a party together and, addressing
both of us, somebody says: “Brown is a member of the libertarian party.” What you will
take away from that utterance is (or at least includes):
(3) somebody x is uniquely such that, a few days ago, I heard x give a competent,
but ultimately not very informative, lecture on the history of psychology;
moreover, x is a member of the libertarian party.
What I will take away from that utterance is (or includes):
(4) somebody x is uniquely such that, a few days ago, x stood next to the maple
tree outside my house; and x was unshaven and generally disheveled; moreo-
ver, x is a member of the libertarian party.
 Chapter 9.  Publicity problems and the nature of linguistic communication 

But even though (3) and (4) are very different, they have the same truth-maker. There
is some x such that each of (3) and (4) is true because the singular proposition x is a
member of the libertarian party is true.
This point generalizes without limit. (We are continuing to confine our attention
to cases where the people in question speak English and are cognitively competent.) If,
ten days later, I say to you: “Brown just bought a new car”, the thought that motivated
my utterance of that sentence will have the same truth-maker as the thought which
that utterance leaves you with. So even though the thoughts that I have in connection
with sentence-tokens of the form FBrown has phiG will always differ in significant ways
from the thoughts that you have in connection with such sentence-tokens, those two
sets of thoughts will be coordinated in a very strict fashion. They will categorically
have the same truth-makers. And this – nothing else – is what is important to our be-
ing able to communicate with one another through language.
Suppose that, in order to communicate with each other, you and I had to associate
exactly the same propositions with utterances of “Brown looks tired”, “there is no way
that Seymour can afford the BMW that he just bought”, and so on. In that case, com-
munication would be impossible. Given the descriptive nature of sense-perception,
and given the public nature of languages like English and Russian, there is no way that
we could possibly have qualitatively identical thoughts in connection with “Brown
looks tired” or, for reasons similar to those just given, in connection with any other
utterance. Communication is possible only because communication does not presup-
pose a complete cognitive convergence between speaker and auditor. For communica-
tion to occur, only a limited convergence is needed: the speaker’s thoughts must have
the same truth-maker as the auditor’s.

Why our analysis accommodates the datum that linguistic communication

is possible

Some would take these remarks to show that communication is impossible.5 Some
would say that, if indeed the proposition I associate with an utterance of “Brown looks
tired” isn’t identical with the proposition that you associate with it, then we simply
aren’t understanding each other: there is only an illusion of communication, fostered
by a kind of parallelism between our thoughts.
Here is my view. The line of thought presented in the last paragraph is either false
or it is a purely terminological point, masquerading as one that is substantive. It is a
datum that people make themselves understood through language. I say to you: “your
mother is coming tomorrow.” Something has been communicated; and that is why you

5. This seems to be what Russell (1917: 152-167) says. That position also seems to be implicit
in Frege’s (1892) view.
 Conceptual Atomism and the Computational Theory of Mind

immediately start tidying your house, washing the bedding in the guest bedroom, and
so forth. Verbal communication is an obvious and pervasive phenomenon.
The naive model of this phenomenon is this: communicating involves the speaker
and the auditor associating exactly the same propositions with the noises that are pro-
duced during their exchange; and so far as they fail to do so, they are not really commu-
nicating – even though they may feel as though they are. As we’ve seen, this position is a
non-starter, given the public nature of linguistic symbols and given the formidable cog-
nitive prerequisites to internalizing and manipulating those symbols. This isn’t to say that
there is never a convergence of the sort just described. But it would be false to say that
communication categorically presupposed such a convergence.
A more realistic model is the one proposed a moment ago. Communicating in-
volves both speaker and auditor associating propositions that have the same truth-
makers with the noises that are produced in during their exchange. Of course, if P and
P* are the same proposition, then they have the same truth-maker. So our model is
compatible with the possibility that, in at least some cases, communicating involves
speaker and auditor associating the very same proposition with a certain sound or ink-
mark. But our model, unlike the naive one, is consistent with the fact that those two
propositions are often distinct. What is necessary, according to our model, is that both
propositions have the same truth-maker and are thus coordinated with each other. In
general, linguistic communication involves a coordination, not an identity, of thought.
Speaker and auditor must have parallel, not necessarily coincident, thoughts.
Because linguistic communication involves identity of truth-maker, and not of
proposition, people are able to communicate with each other despite what may be vast
differences in their personal circumstances.
I have learned much from Aristotle’s words. This wouldn’t be possible if Aristotle’s
communicating his insights to me involved my taking from his words exactly the thoughts
that prompted him to produce them. It wouldn’t be possible if I had to associate with the
word “Plato” exactly the description that Aristotle associated with the corresponding
Greek word. It wouldn’t be possible if I had to associate exactly the same proposition with
the sentence “Plato was a fine philosopher” that Aristotle associated with the correspond-
ing Greek sentence. Given how different my life-history is from Aristotle’s, there is no
possibility of such a complete cognitive convergence.
Language cuts through these biographical differences. This is because linguistic com-
munication is about identity of truth-maker, and not of proposition. It is about what is ex-
ternal, and not about the internal processes through which we learn about what is external.
Of course, not all communication concerns what is external. One can tell people
what one is thinking and feeling; I can tell you that I am happy or sad. But a precondi-
tion for my being able to do this is that linguistic communication not involve the strict
cognitive convergence between speaker and auditor demanded by the naive theory.
 Chapter 9.  Publicity problems and the nature of linguistic communication 

The word “sad” is a public entity. You learn what it means, at least in part, by seeing
how it is used or, conceivably, by having it defined for you.6 You thus learn what it
means through sense-perception. You access that semantic information through the
biography-specific information embodied in your perceptions. Further, having up-
loaded that sensory information, you must then correlate it with your awareness of an
internal state.7 Given any two people who know what “sad” means, each will differ
considerably from the other in regards to the identity of the pre-semantic sensory in-
formation through which he uploads the relevant semantic rule and also in regards to
the nature of the internal state that he correlates with that rule. (You and I learned what
“sad” means through different bodies of perceptual information. Further, you and I
may well have correlated that term with qualitatively different internal states. Surely it
cannot be taken for granted that everyone who knows what “sad” means has exactly
the same sort of sadness. I am not referring to garden-variety skeptical problems about
other minds, but to the fact that people are sad in different ways – just as objects can
be blue in different ways: some are turquoise, some are azure, and so on. Surely I can-
not assume that I have ever experienced a sadness just like any sadness that you have
experienced.)
So when it comes to expressions that report inner-states, it is especially important that
communication involve identity of truth-maker, and not of proposition. For where such
expressions are concerned, there are two barriers to the strict cognitive convergences de-
manded by the naive theory. First there are the inevitable differences in the perceptual
information through which any two people learn what a given word means. Second, there
also are the inevitable (or, at least, probable) differences in the exact phenomenologies of
the inner states that are subsequently correlated with the information provided by those
perceptions. So our model is consistent with the fact that we can communicate our feel-
ings and thoughts, and the competing model is inconsistent with that fact.

Other reasons to accept this analysis

According to the naive view, genuine communication occurs only to the extent speak-
er and auditor associate exactly the same propositions with the expressions used in
their exchange.8 I would now like to develop our criticisms of the naive view.
You and I both speak English. Let S1…Sn be the set of English sentences that you and
I both understand. As we’ve discussed, for practically any value of i, the proposition that

6. Of course, unless you are learning English as a second language, the second method is not
going to be the one that is used.
7. In any case, this is what practically everyone believes. Wittgenstein (1958) famously denies it.
In the last footnote of Chapter 12, we will discuss why Wittgenstein’s view is simply a muddle.
8. I think that (A) entails (B) and that, in fact, they are equivalent. If that is right, then (B) is
technically redundant. But we needn’t investigate this subtlety here.
 Conceptual Atomism and the Computational Theory of Mind

I associate with a token of Si – in other words, the thought I have in connection with such
a token – will be very different from the proposition that you associate with it. At the
same time, the truth-maker of the one thought will be identical with the truth-maker of
the other. So there is some x such that the singular proposition x was wise is the truth-
maker of the proposition I associate with “Socrates was wise” and also of the one that you
associate with it. This means that, even though you and I access the meaning of “Socrates
was wise” through different bodies of information, we are nonetheless in complete agree-
ment as to what state of affairs must hold for that sentence to be true.
Of course, “Socrates was wise” was arbitrarily chosen, and what we just said of it is
true of each sentence in S1…Sn. So what is true of “Socrates was wise” will also be true of
sentences reporting evidence relevant to the truth of that sentence. Suppose recent his-
torical research proves beyond a shadow of a doubt that Socrates died when Plato was
only ten years old. Given our model, there is some x and some y such that the proposi-
tions that both of us associate with “Socrates died when Plato was only ten years old” are
true exactly if the singular proposition x died when y was only ten years old is true.
Thus, given our model of semantic competence, it follows that ceteris paribus you
and I are in complete lock-step in regards to the evidential and, more generally, infer-
ential significance of any sentence of English. A corollary is that the validity of the ar-
guments literally meant by one’s words will perfectly mirror the validity of the argu-
ments that are grasped by the thoughts behind those words. What we just said about
validity is true of truth, cogency, and every other logically significant property.
Our model thus accommodates the fact that sentences of English are capable of
unambiguously transmitting information about the world; and it reconciles that fact
with the fact that any two English speakers will access the meaning of practically any
given English sentence through different bodies of information.
 chapter 10

Content-externalism and self-knowledge

Suppose that, for any truth or falsehood P, the grounds that X has for believing P coin-
cide with the grounds that Y has for believing P. In that case, let us say that X and Y are
in “epistemically identical situations.”
Let W be some counterfactual scenario where your physiological condition is
identical with your actual physiological condition. (So, for any time t, your physiolog-
ical condition in W at t coincides with your actual physiological condition at t.) Sup-
posing that the mental supervenes on the physical, it is obvious that your epistemic
situation in W coincides with your actual epistemic situation. In other words, in W,
your grounds for believing any given proposition coincide with your actual grounds
for believing that same proposition.
According to content-externalism, the thoughts and perceptions that you have in W
might have very different contents from your actual thoughts and perceptions. Given
only that, in actuality, you are thinking Einstein was smart, it doesn’t follow, according to
content-externalism, that in W you are having the same thought – it could be that you
are thinking Laplace was smart or even that you are altogether failing to have a thought.
Given that your epistemic situation in W coincides with your actual epistemic
situation, your actual grounds for believing that you are now thinking Einstein was
smart must be identical with your grounds in W for believing that same proposition.
Here it becomes clear that content-externalism is inconsistent with some basic facts
about the phenomenon of self-knowledge. Your actual grounds for thinking that you
are consciously thinking Einstein was smart are unexceptionable. Given what those
grounds are, you couldn’t possibly be wrong. Your grounds in W for imputing that
same thought to yourself coincide with your actual grounds for doing so. But if con-
tent-externalism is right, you could be wrong in W. It could be that you are thinking
Laplace was smart or that you are altogether failing to have a thought. This means you
could actually be wrong to impute that thought to yourself. This, in turn, means that
you don’t have unexceptionable grounds for believing that you are thinking Einstein
was smart. Since it is a plain fact that you do have such grounds, content-externalism
must be wrong.
Here is a similar argument. If content-externalism is right, you cannot possibly
think Einstein was smart in a world where Einstein doesn’t exist, and you therefore
cannot rightly attribute that thought to yourself in such a world. Whether Einstein
existed obviously cannot be decided through introspection and must be decided on
the basis of historical research. It follows that, if content-externalism is right, one’s
belief that one is consciously thinking Einstein was smart is no more likely to be cor-
 Conceptual Atomism and the Computational Theory of Mind

rect than one’s belief that Einstein existed. No matter how solid one’s grounds are for
believing that Einstein existed, it is always an epistemic possibility that he did not and,
if content-externalism is right, it is therefore an epistemic possibility that one is not
thinking that Einstein was smart. So supposing that content-externalism is right, even
if you think that you are consciously thinking Einstein was smart, it is, from your view-
point, epistemically possible that you are wrong: indeed, it is no less epistemically im-
possible than any other proposition concerning the external world.
But this seems very wrong. One’s beliefs about one’s conscious mental contents are
surely more secure than one’s beliefs about the external world. If I am consciously
thinking that Einstein was smart, it is not, from my viewpoint, an epistemic possibility
that I am not having that thought. In making our beliefs be object-involving with re-
spect to constituents of the external world, content-externalism makes our knowledge
of our own conscious states as uncertain and as lacking in definitive grounds as our
beliefs about the external world. But we have known since Descartes that, while one can
coherently have attitude of skepticism as regards occurrences in the external world, one
cannot coherently have such an attitude as regards occurrences in one’s own conscious-
ness. Because it is inconsistent with this fact, content-externalism must be wrong.
Some content-externalists (e.g. Tyler Burge) have responded by arguing that, de-
spite appearances, content-externalism is consistent with our pre-existing views about
self-knowledge. Others (e.g. Donald Davidson) have argued that, given content-exter-
nalism, those views must be jettisoned. In this chapter, I will argue that
content-externalism is inconsistent with our existing views about self-knowledge and,
further, that those views are correct and therefore mustn’t be jettisoned.

Burge on self-knowledge

Tyler Burge (1986) is a content-externalist who is aware of this problem, and has pro-
duced a brilliant way of dealing with it. His argument is as follows:
(TB) Suppose that you are thinking: I enjoy reading Descartes. Let T be that thought.
Now suppose that you are now thinking that you enjoy reading Descartes. Let T*
be that thought.
T* actually comprises T. So in thinking I am now thinking that I enjoy reading
Descartes, you are necessarily thinking I enjoy reading Descartes. The very occur-
rence of T* thus guarantees its own truth. It is a self-fulfilling thought.
To be maximally clear, let us restate this argument. T* is a thought to the effect
that T is occurring. T is a veritable component of T*. Therefore T* cannot possibly
occur without being correct. Therefore, T* is grounds for its own truth. You can-
not have T* without knowing T*
So if you believe that you are thinking I enjoy reading Descartes, that belief is
grounds for its own truth. You cannot have that belief without having logically adequate
grounds for believing it and, indeed, without believing it on those very grounds.
 Chapter 10.  Content-externalism and self-knowledge 

Obviously this gives a certainty to one’s belief that one is thinking I enjoy read-
ing Descartes that doesn’t attach to one’s believing some proposition about the exter-
nal world. But, in addition, it gives a certainty to that belief that doesn’t even attach
to one’s belief in the principles of logic or mathematics.
This last point is crucial. If you believe that 7+5=12, your belief cannot possi-
bly be wrong. But your thinking 7+5=12 is not what makes that thought true.
7+5=12 is true in virtue of facts about the numbers and arithmetical operations. If
you had to prove to somebody that 7+5=12, it wouldn’t be enough to say: “I just
thought it; therefore its true.” You’d have to go through the sort of rigmarole that
Whitehead and Russell went through. By contrast, if you think I am now entertain-
ing the proposition that I enjoy reading Descartes the grounds for the truth of that
thought are simply that you had it.
For this reason, your thoughts about what you are (consciously) thinking
have the sort of certainty that Kant described as “apodictic.” Not only is it logi-
cally impossible that such thoughts should be wrong. Those occurrences provide
the grounds for their own truth; and, in virtue of having those very thoughts, you
are believing in their truth on those very grounds. So your having those thoughts
guarantees that you recognize the reasons for which they are true and also that you
judge them to be true for those very reasons. So if you believe that you are think-
ing some thought, the certainty that attaches to that belief is of a higher order than
the certainty which attaches to your belief that 7+5=12 or even that triangles have
three sides.
There is one last point to make. The argument just given involves but one
premise: in thinking I believe that I enjoy reading Descartes, one is thinking I enjoy
reading Descartes. We’ve seen that, given that one premise, everything else follows.
It should be pointed out that, in using that premise, the content-externalist is in
no way begging questions against his opponent. The truth of that premise is a
datum. In any case, its truth is known on grounds that are entirely independent of
the truth of content-externalism or of any other doctrine.
So the argument just given is cogent. There is, at most, only an appearance of
conflict between content-externalism and our belief that we can know the identi-
ties of out conscious thoughts with Cartesian certainty.

Let us evaluate this argument.1 Just as Burge says, thinking

T*: I am thinking that I enjoy reading Descartes
does involves thinking
T: I enjoy reading Descartes.

1. I repeat that what I say here is not unlike – though it by no means coincides with – what
McKinsey (1991) and Bilgrami (1992) say.
 Conceptual Atomism and the Computational Theory of Mind

That is undeniable. There is no denying that, if it occurs, T* is necessarily correct. So if

you manage to think T*, your belief is ipso facto correct. And, just as Burge says, the
reason for this is that thinking T is a pre-condition for thinking T*.
But the very thing that is in question is whether content-externalism allows that
pre-condition to be met in the relevant contexts. And it is easily shown that there are
situations where content-externalism doesn’t allow one to have T but where, at the same
time, it is clear on independent grounds that one is having T and where, further, one
knows with Cartesian certainty that one is having T.
Let w be a world where Descartes never existed but that is otherwise just like our
world. (In w, your condition is identical with your condition in our world, leaving
aside the fact that the distal causes of your condition in w are not identical with the
distal causes of your condition here.) Now suppose that content-externalism is correct
and, once again, think (or try to think) to yourself: I enjoy reading Descartes. So far as
your mental state consists in your having that thought, your counterpart in w is failing
to have a thought. Now think: I am thinking that I enjoy reading Descartes. So far as
your mental state consists in your having that thought, your counterpart in w is failing
to have a thought. For your counterpart’s “thought” has a blank in it where your thought
contains the proposition I enjoy reading Descartes.
Supposing, as we are, that content-externalism is correct, your counterpart is
wrong. He attributes a thought to himself that he doesn’t actually have. In the place
where you are actually having a thought your counterpart is shooting a blank. At a
certain juncture in your life, you know with Cartesian certainty that you are having a
certain kind of thought. But if content-externalism is right, your counterpart is not, at
the corresponding juncture in his life, having a thought and is thus wrongly believing
himself to have a thought. Given that your epistemic situation coincides with your
counterpart’s, this is equivalent to saying that you could be the one who is wrong: you
could be mistaking some non-thought of yours for an actual thought. So if content-
externalism is right, your epistemic situation is consistent with your being wrong about
whether you are even having a thought. A fortiori, in direct contradiction to what Burge
maintains, your epistemic situation is consistent with your being wrong about which
thought you are having. (If you don’t know with Cartesian certainty that you are even
having a thought, you don’t know with Cartesian certainty that you are having some
specific thought.)
An immediate consequence of content-externalism is that, even if you were in a
world where your epistemic situation was identical with your actual epistemic situa-
tion, you could still fail to have many of the thoughts that you actually have. In par-
ticular, there are worlds where you cannot think I enjoy reading Descartes but where
your epistemic situation is identical with your epistemic situation in our world. When
we want to know whether content-externalism is compatible with what we know about
the nature of self-knowledge, we have to consider worlds that are epistemically identi-
cal with ours. Therefore, we must not confine our attention to worlds where Descartes
 Chapter 10.  Content-externalism and self-knowledge 

existed. After all, the set of such worlds does not include all of the worlds where your
epistemic situation is exactly as it is here.
Given how Burge and other content-externalists construe what it is to think about
Descartes, Burge’s argument only takes into account the set of worlds where Descartes
exists. It doesn’t take into account the much larger set of worlds where Descartes
doesn’t exist but where your epistemic situation is exactly as it is here. So it doesn’t take
into account all of the worlds in terms of which claims about self-knowledge must be
evaluated, and instead question-beggingly confines itself to those where Descartes has
already been loaded, externalist-style, into your thoughts.
To sum up, depending on how one chooses to look at the matter, Burge’s argument
either begs the question, by restricting its attention to worlds where Descartes has been
made a constituent of our thoughts, or it erroneously identifies the class of worlds that
are epistemically identical with ours with some subset of the class of worlds where
Descartes exists. Either way, the argument is fallacious.

Another content-externalist stratagem

Our position is this:

(CC) If the content-externalist is right, there are worlds epistemically just like
ours where, instead of thinking I want to drink water, I am thinking I want to
drink XYZ. Given that those worlds are epistemically indistinguishable from ours,
it follows (tautologously) that I have no way of knowing that I am not in such a
world and, therefore, that I have no way of knowing (with any special Cartesian
certainty) that I am not thinking I want to drink XYZ as opposed to I want to drink
water. But it is a datum that I do know with Cartesian certainty that, right now, I
am thinking I want to drink water. Given that it is inconsistent with this datum,
content-externalism must be wrong.

We considered Burge’s response to (CC), and found it to be unsatisfactory. But other

content-externalists have a different objection to (CC), to wit:
(SK) In this world, there is water (H2O), not XYZ. For obvious reasons of a generally
logical kind, when you think:

T*: I am thinking that I want to drink some water,

you must also think:

T: I want to drink some water.

Given this, suppose that you were in a world w that was just like our world except
that, in w, there is XYZ instead of H2O. (It follows that your epistemic situation in
w is just like your epistemic situation here.) So, just as you say, instead of thinking
T, your counterpart in w thinks:
 Conceptual Atomism and the Computational Theory of Mind

S: I want to drink some XYZ.

But this doesn’t invalidate his beliefs about what he is thinking. For instead of
thinking T*, your counterpart thinks:

S*: I am thinking that I want to drink some XYZ.

And, of course, S* is correct. So your counterpart is right, and content-external-
ism doesn’t violate his Cartesian access to his own conscious thoughts.
The general principle here is clear. Just as content-externalism says, what you
think is constitutively dependent on facts about your environment. But your
thoughts about your thoughts inherit those same constitutive dependencies. So
your thoughts about your thoughts track your thoughts, ensuring that what you
know with Cartesian certainty in one world coincides exactly with what you know
with Cartesian certainty in worlds epistemically identical with that world.2

(SK) involves a failure to give due attention to distinctions of scope.

Let w be a world where there is XYZ instead of water, but that is otherwise just like
our world. Let w1 be a world where there is X1Y1Z1 instead of water, but that is other-
wise just like our world; let w2 be a world where there is X2Y2Z2 instead of water, but
that is otherwise just like our world; and so on. Off course, your epistemic situation in
our world is identical with your epistemic situation in each of w1, w2, and so on.
As before, let T be the thought: I enjoy drinking water, and let T0 be the thought: I
enjoy drinking XYZ. Further, let T1 be the thought: I enjoy drinking X1Y1Z1; and let T2
be the thought: I enjoy drinking X2Y2Z2; and so on.
Like everyone else, the content-externalist admits that we don’t know with Cartesian
certainty that we are in a water-world, as opposed to an XYZ-world (or an X1Y1Z1-world…).
At the same time, the essence of the content-externalist’s position is this: if I am in w, I am
thinking T0; if I am in w1, I am thinking T1; if I am in w2, I am thinking T2; and so on.
Given this, suppose for the sake of argument that (SK) is cogent. In that case, if I
am in w2, then there is some thought – namely T2 – such that I am having T2 and such
that I am attributing T2 to myself. But notice that, in the last sentence, the “there is” is
given wide-scope with respect to the other operators – with respect to the expressions
“attributing” and “having.” So if (SK) is cogent, then in w there is a thought that I am
having and that I am correctly attributing to myself – but I don’t know which thought it
is. I don’t know whether it is T1 or T2 or T3…
Some fiction may clarify the structure of this argument. Because of some accident
of genetic engineering, there are various people who look (and also dress and talk and
act) exactly the same. These people are John1, John2...and John10. One day somebody
who is in fact John4 is standing in front of me. Under these circumstances, I am in fact
seeing John4 – and not John5 or John6 or any of the other Johns – but I don’t know this
fact. All I know is (at most) that I am seeing either John1 or John2…or John10.

2. This is argued by Kevin Falvey. See Falvey (1994) and Falvey and Owens (1994).
 Chapter 10.  Content-externalism and self-knowledge 

At the same time, there is somebody x such that I know that I am seeing x. After
all, there is some person who is in front of me such that I am seeing him and such that,
being a self-conscious creature, I know myself to be seeing him. (This isn’t a case of
blind-sight – of seeing without knowing that one sees.) What I don’t know is that John4
is that person. So in this scenario, there is somebody x such that x is John4 and such
that I know I am seeing x. At the same time, I don’t know that I am seeing somebody x
such that x is John4. (Another short story will make my meaning clear. Mary and Katy
are identical twins. Mary is standing in front of me, and I see her. But I don’t know
whether the person I am seeing is Mary or her twin. In this case, there is somebody x
such that x is Mary, such that I am seeing x, and such that I know that I am seeing x.
But I don’t know that I am seeing somebody x such that x is Mary. The same principle
applies in the case of John4.)
According to content-externalism, when I am in w2 I am thinking: I enjoy drinking
X2Y2Z2. But my epistemic situation in w2 is just like my epistemic situation here on
Earth. If (SK) is cogent, then there is some thought x such that x is in fact identical with
T2 and such that I know with Cartesian certainty that I am having x. But I don’t know
with Cartesian certainty that there is some thought x such that x is T2 and such that I
am now having T2. The only way I could know that fact with Cartesian certainty would
be to know with Cartesian certainty the chemical structure of the liquid in bathtubs
and swimming pools. But, as the content-externalist himself grants, I cannot know
facts of that kind with that kind of certainty.
So if (SK) is cogent, then I indeed know with Cartesian certainty that I am having
a thought that is in fact T2. But regardless of whether (SK) is cogent, I don’t know with
Cartesian certainty that I am having T2 (as opposed to T3 or T4…) So even if (SK) is
cogent, I don’t know which thought I am having – I don’t know which thought I am
imputing to myself with Cartesian certainty.
Our pre-theoretic intuition is that if I am consciously thinking I enjoy drinking
water, I know with Cartesian certainty that I am thinking I enjoy drinking water. Our
presumption is not given by the claim:
(WS) There is some thought x such that x is identical with I enjoy drinking water and
such that I know with Cartesian certainty that I am thinking x.
Rather, our presumption is given by the claim:
(NS) I know with Cartesian certainty that I am thinking some thought x such that
x is identical with I enjoy drinking water.
So even if (SK) is cogent, that argument fails to show that content-externalism is com-
patible with the fact that we know which thoughts we are having. So far as (SK) ever
appeared to show otherwise, it was because the just discussed ambiguities of scope
were overlooked. Further, for reasons that we’ve already discussed, content-external-
ism couldn’t possibly be consistent with our knowing, with any special certainty, which
thoughts we are consciously having. After all, it could show as much only if it could
 Conceptual Atomism and the Computational Theory of Mind

also show that we had incorrigible knowledge concerning the chemical composition of
water, the authorship of various early modern philosophical texts, and so on. Since we
obviously don’t have such knowledge, content-externalism is false.

Davidson on self-knowledge

Given what it is that we can know about ourselves, and with what degree of certainty
we can know it, content-externalism must be wrong. In any case, we have seen some
good reasons to accept this and no good reasons to reject it. But Donald Davidson re-
jects it, and he provides a bold argument on behalf of his position.3
Davidson’s argument must be understood in terms of an uncontroversial point
about the nature of measurement. From the viewpoint of a physicist, it is irrelevant
whether you measure weight in pounds or kilograms. What matters is that you use
some consistent system of measurement. My weight in pounds is 195. So far as there is
a significant connection between the number 195 and my weight, that reflects the con-
tingent and logically insignificant fact that a certain system of measurement has been
adopted. Given any object O, and given any number N, there is some viable system of
measurement S such that, relative to S, O has mass N. It is only relative to some arbi-
trarily chosen system of measurement that there is any significant connection between
any object’s mass and any number. (By way of anticipation, this premise seems correct.
It is the other premises that we will take issue with.)
In light of these points, suppose that Smith believes that water quenches thirst.
According to Davidson, Smith’s relation to the proposition water quenches thirst is
comparable to the relationship between my body-mass and the number 195. Relative
to one viable system of measurement, there is a special connection between my weight
and the number 195; and relative to others, there isn’t. Relative to one viable system of
description, there is a special connection between Smith’s mental state and the propo-
sition water quenches thirst; and relative to others, there isn’t. Just as it is useful, but not
necessary, to measure weight in pounds, so it is useful, but not necessary, to describe
mental events in terms of propositions.
This brings us to the last step in Davidson’s argument. Statements like “Smith
thinks that there is water, not gin, in the glass” and “Smith wants nothing more than to
own the world’s largest gold mine” characterize Smith’s mental state in terms of Smith-
external states of affairs and, therefore, in terms of categories (gold, water, gin) that
cannot be understood strictly in terms of Smith’s internal, psychological processes.
And it is good that this is so: a language that described Smith’s internal processes in
terms of categories that had no counterparts in the external world would be quite use-
less in the way of making Smith’s behavior predictable and understandable.

3. Davidson (2005: 15-38). A viewpoint similar to Davidson’s is found in Churchland (1981:

602-603).
 Chapter 10.  Content-externalism and self-knowledge 

Before we close the argument, let us sum up what we’ve just said. (1) A person’s
mental content no more has to be described in terms of propositions than length has
to be measured in yards. (2) When we do ascribe propositional attitudes to people we
are characterizing their mental states in terms of extra-mental realities.
If (1) and (2) are correct, there is no reason to suppose that Smith must, or even
could, have incorrigible knowledge as to what propositional attitudes he has. For, sup-
posing that (1) and (2) are true, ascriptions of propositional attitudes reflect not only
the internal psychological condition of the relevant person, but also facts relating to
mind-independent externalities. Thus, given only what is going on “in Smith’s head”,
there is no reason to say that Smith believes that water quenches thirst. After all, Smith
doesn’t have incorrigible knowledge about what is going on in his environment or
therefore about the physical structures of its constituents. Thus, even if Smith has a
complete knowledge of what is going on in his head, he doesn’t necessarily know that
he believes that water quenches thirst.
In conclusion, given a correct understanding of statements like “Smith believes
that water quenches thirst”, it is clear that we don’t have any special knowledge con-
cerning our own mental states. Content-externalism is consistent with this, and should
therefore be commended, not rejected.

Evaluating Davidson’s argument

Contrary to what Davidson supposes, Smith’s relation to the proposition water quench-
es thirst is not comparable to the relationship between my weight and the number 195.
In the one case, we are dealing with an arbitrary convention (or, more accurately, with
a non-arbitrary consequence of an arbitrary convention). In the other, we are dealing
with a fact that has nothing to do with convention. Not many authors deny this. But
even though, for this reason, Davidson’s view about self-knowledge is not widely ac-
cepted, it is worthwhile to state explicitly why that view is wrong and what is wrong
with his argument for it.
People are usefully described as having attitudes towards propositions. Why isn’t
the same true of rocks and trees? The most obvious answer is: “rocks and trees don’t
have attitudes towards propositions, whereas people do have such attitudes.” But this is
exactly the position that Davidson is attempting to refute, and he is therefore barred
from endorsing it.
And there is much to be said for that position. If you subtract the representational
from the mental, not much is left. Indeed, according to one school of thought, known
as “intentionalism”, nothing mental is non-representational. Even phenomenal content
is representational: given a difference in phenomenology, there is a difference in repre-
sentational content.
 Conceptual Atomism and the Computational Theory of Mind

I myself think that this is an overstatement. For reasons that I have given else-
where4, I think that there are mental entities that are either non-representational or
only derivatively representational. But what is indisputable is that representationality
is a pervasive feature of the mental. It is also indisputable that non-derivative represen-
tationality is a sufficient, if not a necessary, condition for mentality.
Supposing that Davidson is right, there must be some way of identifying and de-
scribing mental phenomena that does not involve the concepts of belief, desire, and so
on. But this doesn’t seem to be the case. Given that the concepts mental and non-de-
rivatively representational are virtually interchangeable, it makes questionable sense to
say that there could be a viable description of psychological reality that didn’t use con-
cepts such as representation and proposition. Where human subjects are concerned,
there doesn’t seem to be a neutral subject-matter that we may or may not choose to
describe in terms of concepts like desire, fear, and so on.5 Davidson’s argument as-
sumes otherwise and is therefore fallacious.
One might make the following objection to this last point:
When we use terms like “belief ” and “doubt”, we are describing what are in fact
neural events in mentalistic terms; and when we use terms like “glial cell” and “c-
fiber stimulation”, we are discussing those same events in non-mentalistic terms.
So contrary to what you say, there is a neutral subject-matter that may or may not
be described in mentalistic terms.

In fact, Davidson himself makes this very point in his essay “Mental Events.”6 There
Davidson says that “events are mental only as described.”7 The idea is that, just as a
person may be described as a lawyer or father, so an event may be described as a pat-
tern of electrical stimulation or as a surge of fear. This seems to entail that psychologi-
cal statements organize data that could just as well be organized in completely non-
psychological terms. It also seems to entail, or at least to suggest, that statements
involving attributions of propositional attitudes (statements like “Smith fears what will
happen to him if he becomes a lawyer” and “I felt a surge of joy at learning that I would
once again be with Mary”) organize data that could just as well be organized by state-

4. Kuczynski (2003).
5. In any case, there doesn’t seem to be any way of describing the contents of the personal
stratum of cognition except in terms of those concepts. It is another matter altogether whether
the same is true of the contents of the subpersonal stratum. In fact, it seems that those processes
are not to be understood in terms of those concepts.
But this doesn’t help Davidson. First of all, Davidson’s point concerns the contents of the per-
sonal stratum of mentation; and we’ve already seen why that point is false with respect to those
contents. It should be noted that Davidson rejects the very idea of subpersonal mentation.
6. Davidson (1980: Chapter 11).
7. Davidson (1980: 215).
 Chapter 10.  Content-externalism and self-knowledge 

ments that don’t involve attributions of propositional attitudes. And this, of course, is
exactly the view that underlies Davidson’s analysis of self-knowledge.
But this line of thought is misguided, as a story may help show. I am suddenly
overwhelmed by a number of complex feelings. I attempt to articulate them. (I say “I
now realize that I should have become an actor, and that my decision to work on Wall
Street – though socially respectable and seemingly indicative of a mature acceptance
of life’s exigencies – was ultimately an attempt to avoid confronting the question that
has been haunting me since the age of five, to wit…”) Let S be the statement that re-
sults. Let us suppose that S is accurate.
Later that day, I am watching a film of the brain-event that realized the just-de-
scribed mental event. (That brain-event was recorded by a micro-camera inside my
cranium.) I describe what I see, using the appropriate neuro-physiological terms. Let
S* be the statement that results. Let us suppose that S* is accurate.
The event in virtue of which S is true is identical with the event in virtue of which
S* is true. But the data of which S is a correct synthesis is entirely different from the
data of which S* is a correct synthesis. S* conceptualizes various visual perceptions of
various discolorations on a certain video-monitor. That body of data doesn’t even
overlap with the body of data conceptualized by S. In general, there is no neutral data
of which both psychological and non-psychological syntheses are possible.
Another illustration of this principle may be in order. A pang of remorse may be
identical with some series of neural events. But the data correctly synthesized by a verbal
expression of remorse is entirely different from the data correctly synthesized by any
statement of the form Fneural events of type PHI are now occurring.G Even if I have un-
exceptionable grounds for saying “I now feel terrible remorse over stealing my sister’s
cookie”, I may have no grounds for making any statement involving terms like “neuron”
and “glial cell.” So statements of the first kind do not organize data that is also correctly
organized by neurological statements or, by obvious extensions of this reasoning, by
non-psychological statements of any kind.
Given only that mental events are realized by neurological events, it doesn’t follow
that the data organized by mentalistic statements can be organized by neurological or
otherwise non-mentalistic statements. A failure to see this underlies Davidson’s view
that psychological statements provide a way of organizing data that could in principle
be organized along completely non-psychological lines.
This failure is indicative of a confusion that pervades Davidson’s analysis. Obvi-
ously Davidson is right to distinguish between the thing studied and our methods of
studying it. We may use a blue tincture to stain some organelle, so as to make it easier to
observe; and it would obviously be foolish to conclude on that account that the or-
ganelle was really blue. But concepts are not instruments. When I use a magnifying glass
to look at an insect, I am interested in the insect, not the magnifying glass. The magnify-
ing glass helps me generate the relevant data; but no fact about the magnifying glass is
 Conceptual Atomism and the Computational Theory of Mind

itself a constituent of that data.8 By contrast, when I talk about Bill’s fear of heights, I
am not using the concept Bill’s fear of heights to identify some datum that lies on the
other side of that concept. The concept is a constituent of the data.
To be sure, the concepts that we use are not always constitutive of the data being
studied. The phenomenon of heat is understood in terms of the concept molecular mo-
tion. But the latter concept is not constitutive of the data that a theory of heat must model.
That is why one can grasp the concept of heat without grasping the concept of molecular
motion. But the concepts that Davidson is discussing – belief, intention, doubt, and so on
– are constitutive of the data that a psychological theory must model. I may or may not
choose to understand Bill’s aversion to public places in terms of the concepts excessively
punitive super-ego and conversion reaction. But if I eschew the concept fear of being in a
public place, I obliterate the very data that I am trying to understand.9

8. Facts about the magnifying glass may be meta-data. If it turns out that magnifying power
of the glass was less than what I was led to believe by the person who sold it to me, or it turns out
that our views about optics are seriously wrong, then I will re-appraise the data that I generated
with the help of the magnifying glass. I will, for example, revise my view as to the size of the
dust-mite that I was studying. But this shows that the magnifying glass is of interest to the ento-
mologist only to the extent that it helps him generate data about insects. Information about the
magnifying glass itself is of interest only in so far as it is relevant to the generation of such data.
9. Some philosophers, most notably Paul Churchland, deny that mental phenomena are to be
understood in terms of concepts like belief, doubt, and the like. Such concepts, argues Chur-
chland, belong to folk-psychology, and are to be given no more credence than the concepts of
folk-physics or folk-medicine. Given this, it might seem that Davidson’s position is immune to
the arguments we’ve put forth, provided that Churchland’s view is right.
This is not the case. Churchland doesn’t hold that representational content doesn’t exist, but
rather that it has been misunderstood. In fact, Churchland’s view is that, given a system of folk-
psychological categories, it is not possible to account for the fact that so much information (or
content) is housed and processed by brains. Further, so far as there is support for Churchland’s
rejection of folk-psychology, that support lies in the success of theories (such as those of Chomsky
and Marr) that very much take for granted the existence (and efficacy) of mental content. Indeed,
those theories posit extensions, not diminutions, of the range of phenomena that are to be unders-
tood in terms of the processing of representational content. So no support for Davidson’s position
is found in Churchland’s position or in anything that supports Churchland’s position.
Churchland describes himself as having an “eliminativist” attitude towards the mental. But
really Churchland has an eliminativist attitude not towards the mental, but towards the entities
posited by what he refers to as folk-psychology. Churchland has a strongly realistic attitude to-
wards subpersonal mental processes. Indeed, as we just noted, his eliminitavism in regards to
folk-psychological entities is grounded in his acceptance of mental entities that defy folk-psy-
chological description. See Churchland (1981: Section V). In general, so far as eliminitavism
(so-called) is plausible, the arguments for it presuppose the acceptance of mental representa-
tions, albeit ones that are alien to our ordinary way of thinking about the mental. So Davidson’s
position gains no support from the eliminitavist’s.
This is not to mention that much of Davidson’s work implodes if theories positing subper-
sonal thought are correct. (For example, Davidson argues that only creatures that already speak
 Chapter 10.  Content-externalism and self-knowledge 

We’ve seen that Premise (1) of Davidson’s argument is false. Contrary to what
Davidson holds, my relationship to the propositions that I believe, desire to be true,
and so on, is not comparable to the relationship that my weight bears to the number
195. I really do believe that two is an even prime, and I really do doubt whether others
will accept my analysis of definite descriptions. It isn’t simply that my mental states are
usefully put into some kind of correspondence with the propositions two is an even
prime and others will accept my analysis of definite descriptions.
But what about Premise (2)? Surely Davidson is right to say that we often charac-
terize a person’s mental states in terms of states of affairs that are external to him and
whose constitutions may therefore be unknown to him. And Davidson is right to say
that, because we do so, people are better predicted and understood than they would
otherwise be. At the end of Chapter 2 we ourselves said why this is so. (I leave it open
to what extent our reasons for embracing this unorthodox conclusion coincide with
Davidson’s.) Language would be unwieldy at best, useless at worst, if it described peo-
ple in terms of contents that they actually grasped. Given only what my eyes are telling
me, I could be seeing either Mary or Twin-Mary. But I know that Twin-Mary is in
Finland, that I must therefore be seeing Mary and, consequently, and that I can truly
say “I see Mary.” If I confined my reportage to what my visual perception, by itself, was
telling me, I would flood my auditor with useless and tedious information. If I said “I
had a perception to the effect that, in such and such a location, there was an object x
such that x had morphology M and such that...”, I would not tell you what you wanted
to know, and I would tell you what you didn’t want to know. (Nobody other than a few
art teachers and experimental psychologists wants to know what my visual perception,
by itself, is telling me. People want to know whether I saw Mary (as opposed to Twin-
Mary or Fred or...). They care whether I saw a person with such and such characteristics
only to the extent it makes it likely that I saw Mary.) So a misrepresentation of the
content of my visual perception is not only useful but positively necessary.
This point generalizes. Consider the sentences: “Bill thinks that Hesperus is lovely”,
“Jerry wants to heat the water in his pool”, and “Larry hopes that you cut the grass.” We
saw in Chapters 2–4 that the contents of Bill’s, Jerry’s, and Larry’s mental states are not
correctly represented by those sentences. (But we also saw that sentences that do cor-
rectly represent those contents can be put into a systematic correspondence with the
sentences that we typically use to report them – with, for example, sentences like “Jerry

a language can think. If there are subpersonal cognitive processes, they obviously pre-date the
individual’s ability to use language, given that they would be prerequisites to the perceptions and
cognitive operations involved in the learning of language. So if there is subpersonal mentation,
then Davidson’s views on the relation between language and thought are false. Given how im-
portant those views are to the philosophical system that Davidson produced, that system collap-
ses if there are in fact subpersonal cognitive processes of the kind posited by Chomsky and Marr. )
So Davidson is barred from using Churchland’s view as a way of combating our criticisms, even
though others might not be thus barred.
 Conceptual Atomism and the Computational Theory of Mind

wants to heat the water in his pool” and “Larry hopes that you cut the grass.” Davidson
overlooks this point, and we will soon see why this oversight vitiates his argument.)
Davidson is therefore right to say that it is necessary (or at least very useful) to
describe people as believing (or desiring true or doubting...) propositions that they do
not in fact believe (or wish to be true or doubt...). And Davidson is therefore right to
say that the propositions in terms of which people’s mental contents are fruitfully de-
scribed do not (at least not always, or even often) coincide with those contents.
But Davidson is not right to infer from this that the contents of people’s mental
states don’t coincide with any propositions. Although Davidson is right to say that “Bill
thinks that Hesperus is lovely” does not say what the actual content of Bill’s mental state
is, he is wrong to say that no sentence does so. We’ve seen that Bill’s mental content is
given by some sentence along the lines of: “Bill believes that something x is uniquely a
last celestial body to disappear from the morning sky and Bill also believes that ‘Hespe-
rus’ names x and that x is lovely.” So Bill’s mental state is non-conventionally related to
the proposition encoded in that sentence. Also, we saw that the reason that Bill is use-
fully described by sentences like “Bill thinks that Hesperus is lovely” is precisely that,
although the propositions meant by such sentences are very different from the proposi-
tions that coincide with the contents of Bill’s mental states, the two sets of propositions
can be put into a systematic correspondence.
Premise (2) of Davidson’s argument is correct. (In any case, it is nearly correct. For,
as we just saw, there are some sentences that do describe the actual contents of people’s
mental states. But those sentences are not often used.) And Davidson evidently in-
ferred from this that (1) must be true. But (1) does not follow from (2), and (1) is also
false. Therefore Davidson’s argument is fallacious.

10. It seems to me that underlying Davidson’s analysis is an underestimation of how much two
empirically equivalent systems of measurement must have in common.
Davidson is obviously right to say that, for any object O and any number N, there is some
viable system of measurement S such that, relative to S, O’s mass is N. But it doesn’t follow that
the concepts used by an adequate system of measurement can be jettisoned. In actuality, given
two empirically adequate systems of measurement, it must be possible to establish a strict cor-
respondence between the concepts used by the one system and those used by the other. What is
true of measurement is true of description generally. Supposing that folk-psychology is empiri-
cally adequate, if only for a small subset of mental phenomena, it follows that concepts similar
to those used by it are indispensable to psychological explanation. To substantiate these points,
we must briefly set aside any discussion of propositional attitudes and focus exclusively on the
concept of measurement.
In some cases, the differences between systems of measurement are purely verbal. Length
can be measured in feet or in yards. If we are measuring an object’s length in feet, let us say that
we are using system MF; and if we are measuring it in yards, let us say that we are using MY. The
statement:
(i) “X is three feet long”
is only verbally different from
 Chapter 10.  Content-externalism and self-knowledge 

(ii) “X is a yard long.”

(i) and (ii) use the very same concepts, and they differ only in respect of the words they use to
refer to X’s length. In general, MF and MY use the same concepts, and differ only in respect of the
expressions they use to express those concepts.
Not all differences between systems of measurement are as trivial as the one just described. If
the distance between x and y is small (but not too small), it can be measured in terms of how many
rods of a given length are needed to connect x and y. But distances of considerable length, e.g.
those between celestial bodies, cannot be measured with yard-sticks. For this reason, such distan-
ces must be measured in other ways, e.g. in terms of the amount of time it takes for light to travel
from x to y. Here we are dealing with two genuinely different systems of measurement.
But great care had to be taken to ensure that, where distances that can be measured by
either method are concerned, the measurements generated with the one system coincide with
those generated by the other. Otherwise, the statement “the distance between the Earth and the
Moon is over a thousand times greater than the distance between Santa Barbara and Los Ange-
les” would be meaningless (or ambiguous). For exactly similar reasons, where moderate tempe-
ratures (such as those on the surface of the Earth) are concerned, the methods that are used to
measure the temperature of a very hot body, such as the sun, must yield the same results as
those generated by using a mercury-thermometer. Otherwise statements comparing the tempe-
rature of Santa Barbara with those on the surface of the Sun would be meaningless (or ambi-
guous). Hempel (1952: 62-78, 1965: 123-134) cogently argues for these very points.
For argument’s sake, suppose that there were no way of establishing a one-one correspon-
dence between the amount of time needed for light to travel between two points and the number
of times a meter-rod of a given length could be laid endwise between those points. In that case,
those two methods would measure different properties. Regardless of whether we use meter-
rods or light beams, we can speak of length simpliciter only because empirical research has esta-
blished that (in contexts where both methods can be used) there is a one-one correspondence
between statements reporting the results of the one method of measurement and statements
reporting the results of the other kind of measurement. See Hempel (1952: 62-78, 1965: 123-
134) and Pap (1962: Chapter 8).
In general, a given quantity can be measured in different ways only if it is established that
those different methods yield identical results (with respect to situations to which both methods
are applicable). If two systems of measurement don’t satisfy this requirement, then either they
don’t measure the same quantity or one of them is generating erroneous results.
Before we take the next step in our argument, let us synthesize what we’ve said so far. In some
cases, two systems of measurement are only verbally different. (This is the case with MF and MY.)
In other cases, the equivalence of two systems presupposes a posteriori knowledge of physical law.
But given the equivalence of two such systems, the statements generated by the one can be mapped
onto those made by the other. It is not analytic that (for small distances) operations with meter-
rods yield statements that are equivalent with those yielded by measurements of time-intervals
between light-signals. But given that equivalence, any statement S generated by the one method
will be analytically equivalent with some statement S* generated by the other (provided that S and
S* concern distances that can be measured using either method). So given two adequate methods
of measuring a given quantity, the statements produced by the one method are analytic transfor-
mations of those produced by the other.
 Conceptual Atomism and the Computational Theory of Mind

Why content-externalism should accept that it is inconsistent with our pre-

theoretic intuitions about self-knowledge
It is ill-advised for proponents of content-externalism to attempt to reconcile that doc-
trine with our intuitions about self-knowledge. This is because content-externalism
nearly enough is a denial that such intuitions are correct.
I look at Venus in the morning and sincerely say “that is lovely.” I look at Venus in
the evening and sincerely say the same thing. But I don’t know that I was looking at the
same thing on both occasions.
The content-externalist says that there is some one proposition such that I was
sincerely affirming that proposition on both occasions. More explicitly, there is some x
such that, on each occasion, I believed and affirmed the proposition: x is lovely. It is a
datum – one that content-externalists don’t deny – that, under the circumstances de-
scribed, I need not know this last fact. So the essence of content-externalism nearly
enough is that one doesn’t know what one is thinking. It is therefore not an option for
content-externalists to try to show that their doctrine validates our intuitions about
self-knowledge.
We can corroborate this by modifying the story just told. Consider a scenario that
is just like the one just described, except in this one respect: in the evening, I sincerely
say “that is not lovely.” According to this content-externalist, there is some x such that
in the morning I believe the proposition x is lovely, and such that in the evening I be-
lieve the proposition x is not lovely.
It is a datum that, given only what we’ve just said, I am not necessarily guilty of irra-
tionality – I am surely not to be compared with someone who sincerely says “I believe that
Smith is French and that Smith is not French.” Most content-externalists admit this.11
At the same time, it is clear that I would be guilty of irrationality if I knew that, for
some x, I was sincerely affirming x is lovely and also x is not lovely. So if content-exter-
nalism says that I know with Cartesian certainty what it is that I believe, then that
doctrine is forced to part with the datum that I can coherently affirm both “that [ac-
companied by a morning-ostension of Venus] is lovely” and also “that [accompanied

In light of these points, let us once again look at Davidson’s analysis. Let H be the method
of psychological description that we ordinarily use. So H is the method that generates statements
like Sally believes that 1+1=2 and Fred wants to go to Hawaii.
First of all, it seems to me that H is empirically adequate – at least within its narrow horizons.
It is probably not appropriate for the description of subpersonal cognitive processes. But it is at
least partially adequate for the description of personal psychological processes. It is hard to be-
lieve that statements such as Sally believes that 1+1=2 are always simply wrong (or meaningless).
Let H* be any method of describing the mind that can recognize the obvious fact that Sally
believes that 1+1=2. Given plausible extensions of the points just made regarding measurement,
H* must use concepts that are identical with, or comparable to, those used by H. If this is right,
then concepts similar to those used by H are constitutive of any empirically adequate way of
describing psychological facts.
11. Nathan Salmon (1986) denies it.
 Chapter 10.  Content-externalism and self-knowledge 

by an evening-ostension of Venus] is not lovely.” So it is in that doctrine’s interest that

our pre-theoretic claims about self-knowledge be wrong.
Also, although I myself believe that content-externalism is incorrect, I don’t think
that, by itself, its irreconcilability with our pre-theoretic claims about self-knowledge
is particularly to its discredit. Content-externalism says that there can be cognitive
hallucinations in much the same way that there can be perceptual hallucinations. If the
man I am “seeing” doesn’t exist, then I am not seeing a man and am, to that extent,
failing to have a perception at all. Perception decomposes into a psychological and
non-psychological component: there is no perception if either is lacking. Nobody re-
gards it as puzzling that I can be wrong about who or what I am seeing. I think that I
am seeing Fred when I am actually seeing his twin. The fact that one can be wrong
about the identity of what one is seeing is a straightforward corollary of the fact that
perceptions have a non-psychological component.
Content-externalism plausibly (though incorrectly) says that, in a respect similar
to that just described, thoughts are like perceptions. A thought has both a psychologi-
cal and a non-psychological component. If either is lacking, there is no thought. A
corollary is that one can be wrong about the identity of what one is thinking. In any
case, if we deny that this is a corollary, then we destroy the parallel with perception that
forms the heart of content-externalism.
So the content-externalist’s best bet is to say that “think” is in the same category as
“perceive” and, consequently, that one is no more authoritative with respect to the iden-
tity of what one thinks than one is with respect to the identity of what one perceives.
 chapter 11

Why one’s mental content is fixed by one’s

epistemic situation

If content-externalism is right, then two people who are atom-for-atom duplicates,

and (supposing that materialism is correct) are therefore in epistemically identical
situations, can have thoughts with different contents. I am (let us suppose) having a
thought that has Smith as a constituent. But it is epistemically possible that I am having
a thought that instead has Twin-Smith as a constituent.
In this chapter, I wish to argue that this is incoherent. To say that it is epistemically pos-
sible that I am thinking about Twin-Smith is to say that, given only the data at my immedi-
ate disposal, it cannot be ruled out that I am thinking about Twin-Smith. To say that this isn’t
ruled out by the data at my immediate disposal is to say that it isn’t ruled out by my mental
content. So two people are in epistemically identical situations just in case their mental states
have the same contents. Content-externalism is therefore wrong, given that it demands that
one’s epistemic situation not fix the contents of one’s mental states. I would now like to de-
velop the argument just outlined.
The essence of content-externalism is this:
(1) Given two people who are exactly the same, leaving aside facts about their
respective environments, what the one believes may differ from what the oth-
er believes.
This entails:
(2) Given two people who are in different environments, what the one believes
may differ from what the other believes, even if the two are in identical epis-
temic situations.
X and Y are in identical epistemic situations iff, for any proposition P, the grounds that
X has for believing P are identical with the grounds that Y has for believing P.
(Actually, it would be more accurate to say that X and Y are in identical epistemic
situations iff, for any truth or falsehood P, the grounds that X has for believing P are
identical with the grounds that Y has for believing P. This definition, unlike the previous
one, doesn’t involve the controversial view that all truths and falsehoods are identical
with propositions – a view that we will examine at length in chapters 22 and 23. But, in
this context, solely for the purposes of brevity, we will use the term “proposition” as a
shorthand for “truth or falsehood” or “piece of information.”)
 Conceptual Atomism and the Computational Theory of Mind

X has the same grounds as Y for believing proposition P iff what X already believes
has the same bearing on P as what Y already believes. For you to have good grounds
for believing that somebody stole your car is for the propositions in which you already
believe to have a certain relation one of confirmation to the proposition that somebody
stole your car. So X and Y are in identical epistemic situations exactly if, for any prop-
osition P, the propositions in which X already believes stand in precisely the same re-
lationship to P as the propositions in which Y already believes. Thus, (1) entails:
(3) Given two people who are in different environments, what the one believes
may differ from what the other believes, even if, for any proposition P, the
propositions in which the one already believes stand in precisely the same
relationship to P as the propositions in which the other already believes.
In order for the propositions that X believes to bear on any proposition P in precisely
the same way as the propositions that Y already believes, it is necessary that X and Y
believe precisely the same propositions.
Before closing the argument, there is a fact relating to the concept of analyticity
that we must point out. Even if the propositions that X believes are analytically equiva-
lent with the ones that Y believes, there may still be some proposition P such that the
exact way in which the former bear on P differs from the way in which the latter do so.
Suppose that my one belief is that 1+1=2 and that your one belief is that each thing is
self-identical. In that case, the proposition that I believe to be true entails, and is en-
tailed by, exactly the same propositions as the proposition that you believe to be true.
But the exact way in which the one proposition that I believe bears on for some n,
n+1=2 is obviously very different from the exact way in which the one proposition that
you believe bears on for some n, n+1=2. A derivation of for some n, n+1=2 from the one
proposition, but not the other, will be extremely straightforward. In the one case, but
not the other, a single application of the rule of existential generalization will suffice.
Given what was said in the last paragraph, it follows that (3), and therefore (1), entail:
(4) Given two people who are in different environments, what the one believes
may differ from what the other believes, even if the propositions in which the
one already believes are identical with the propositions in which the other al-
ready believes.
But (4) is a blatant self-contradiction. It says that two people who believe the very same
thing may not believe the very same thing. So content-externalism is given by a self-
contradictory proposition and is therefore false.
To break this argument, the content-externalism would have to go down one of
two paths. He must deny either:
(A) For X and Y to be in epistemically identical situations is for X and Y to have
the same beliefs,
 Chapter 11.  Why one’s mental content is fixed by one’s epistemic situation 

or
(B) Given only that X and Y are atom-for-atom duplicates, and that the mental
supervenes on the physical, it follows that they are in epistemically identical
situations.
Let us start by defending (B). Supposing that X and Y are atom for atom duplicates and
that the mental supervenes on the physical, X has no better reason than Y for believing
that he is in an H2O world and Y has no better reason than X for believing that he is in
an ABC-world. Indeed, neither has any reason not equally had by the other for believing
anything. And this surely entails that they are in epistemically identical situations.
Here is a similar argument. As it happens, you are in an H2O world. But imagine a
counterfactual scenario w such that, in w, you are in an ABC-world, but are otherwise
exactly as you are now. So, in w, you believe that you are in an H2O world. Of course, that
belief is wrong. But is it less rational than your actual belief that you are in such a world?
Of course not. Your sensory surfaces are disturbed in exactly the same way as those of
your counterpart in the ABC-world. The one set of disturbances therefore cannot pos-
sibly differ from the other in respect of the extent to which it warrants a given belief.
Indeed, the two sets of disturbances coincide, not only in respect of how much weight
they confer on any given proposition, but also in the exact manner in which they do so:
in both cases, confirmation travels along the very same channels. This shows that you
have no conceivable grounds for discriminating between your epistemic situation and
that of any one of your atom-for-atom duplicates. In general, for all one knows, one’s
condition is identical with that of any one of one’s atom-for-atom duplicates, and it
follows trivially that one’s epistemic situation coincides with that of each of those du-
plicates. It is therefore not an option to reject (B).
But have we begged questions in assuming (A)? The essence of content-external-
ism is that, even if X and Y are atom-for-atom duplicates and (supposing that material-
ism is correct) are therefore in epistemically identical situations, differences in their
environments amount to differences in what they believe. So, it might be said, our use
of (A) begs the question.
his is not so. Let X be some actual human, and let Y be X’s duplicate in a world that
contains ABC instead of H2O. First of all, the content-externalist cannot plausibly say
that X and Y differ in all of their beliefs. If X thinks 1+1=2, so does Y. Even if content-
externalism is correct, they differ at most with respect to a certain sub-class of their
beliefs. More precisely, their beliefs diverge only when those beliefs are object-involv-
ing with respect to those features of their environments that are not common to both
worlds. Since not every belief that one has is object-involving with respect to one’s
environment, X and Y share some beliefs.
What is the nature of these shared beliefs? Every pro-externalist argument begins
by asking the reader to make a supposition similar to, if not identical with, the follow-
ing: “Let w be a possible world that is just like ours except that, in w, the relatively clear
liquid that comes out of faucets and flows in rivers…is ABC, not H2O.” This supposi-
 Conceptual Atomism and the Computational Theory of Mind

tion is, at least by implication, to the effect that you and your counterpart in w both
have a certain existential belief in common, namely: there is some relatively clear liquid
that comes out of faucets and flows in rivers and...
In general, there is no way to motivate any pro-externalist argument without imput-
ing a common core of existential belief to the doppelgangers in question. Consider Burge’s
classic (1979) argument. In Smith’s world, “arthritis” refers to a disease of the joints. In
Smith*’s world, it refers to some other disease, namely arthritis*, that doesn’t affect joints
alone, but is otherwise just like arthritis. Apart from this difference, Smith and Smith* are
exactly the same. Burge concludes that, if Smith believes aspirin relieves the pain caused by
arthritis, Smith* believes aspirin relieves the pain caused by arthritis*.
As Burge himself clearly sees, if this argument is to be cogent, it must be assumed
that Smith and Smith* are both cognitively normal, linguistically competent, and so
on. In order for somebody to know what an expression means, he must be in com-
mand of a rich body of knowledge. If it is a syntactically simple expression, he must
have encountered occurrences of that term, and must therefore know a great deal
about his environment. The same is true if that expression is syntactically complex,
since one cannot understand a syntactically complex expression without understand-
ing the simple expressions composing it. Thus, each of Smith and Smith* must have
considerable stores of environmental knowledge if he is to be capable of being told, or
otherwise learning, what (in his world) “arthritis” refers to. So, because Burge’s argu-
ment assumes that each person does know what (in his world) is meant by “arthritis”,
that argument goes through only if it is assumed that each has vast stores of knowledge
about his environment.
Further, the one person’s pre-existing knowledge must coincide, or at least overlap to
a high degree, with the other’s. Nothing resembling Burge’s thought-experiment is pos-
sible if either one of Smith or Smith* doesn’t know that, in his world, there is some dis-
ease x such that x causes pain one’s joints and “arthritis” refers to x. In general, Burge’s
experiment presupposes that the descriptive beliefs through which Smith locks onto ar-
thritis coincide (to a high degree) with the descriptive beliefs through which Smith* locks
onto arthritis*. So in assuming, as he must, that Smith’s epistemic situation is identical
with Smith*’s, Burge is demanding that what Smith believes be identical with what
Smith* believes. So Burge’s argument requires the truth of (A). The same is true of all
pro-externalist arguments, and we thus haven’t begged any questions in assuming (A).
In response, it could be said that Burge is assuming only that, by virtue of being in
epistemically identical situations, Smith and Smith* have the same purely existential be-
liefs, it being left open whether they may diverge in respect of their object-involving be-
liefs. But it is clear that, if indeed Smith (Smith*) has any object-involving beliefs, the
contents of those beliefs are fixed by the contents of the strictly existential beliefs that
Smith (Smith*) has. (In any case, we established in Chapters 1-4 that two people cannot
differ with respect to any of their beliefs about the spatiotemporal world unless they dif-
fer with respect to their strictly existential beliefs about it.) Given that Smith and Smith*
have precisely the same purely existential beliefs – given, indeed, that Burge’s thought-
 Chapter 11.  Why one’s mental content is fixed by one’s epistemic situation 

experiment requires this – it follows that Burge is incoherently demanding that two peo-
ple can have different beliefs in virtue of having the very same beliefs.
We may conclude that content-externalism is given by a self-contradictory proposi-
tion: what X believes may differ from what Y believes even if X and Y are in epistemically
identical situations, i.e. even if what X believes is identical with what Y believes. In other
words, people who have the same beliefs may not have the same beliefs.
Let us develop this line of thought by extending Burge’s thought experiment. Smith
knows that he has arthritis but is otherwise in perfect health, and Smith* knows that
he has arthritis* but is otherwise in perfect health. Also, each of Smith and Smith*
knows himself to be the only person in Cleveland (i.e., in his world’s version of Cleve-
land) whose phone number has four consecutive digits in it.
In light of this, suppose that Burge is right. In that case, there is some x such that
x is a disease that affects only the joints, and Smith has the belief:
(S) there is exactly one person in Cleveland whose phone number has four con-
secutive digits in it, and that person has x and has no other diseases.
And there is some y such that y is a disease that doesn’t affect only the joints, and
Smith* has the belief:
(S*) there is exactly one person in Cleveland whose phone number has four con-
secutive digits in it, and that person has y and has no other diseases.
S and S* are incompatible. There is no possible world where they are both true. But
apart from his belief in S, Smith’s epistemic situation is, by hypothesis, identical with
Smith*’s. So the beliefs that Smith has apart from his belief in S are identical in content
with those which Smith* has apart from his belief S*.
As we’ve discussed, Smith believes S in virtue of his having those other beliefs, the
same being true mutatis mutandis of Smith*. For as we just saw, Smith cannot lock onto
the concept of arthritis except by way of the right existential beliefs; and, as we’ve also-
seen, Smith* will have those very same existential beliefs. So if, as Burge supposes, Smith
believes S, that is in virtue of the fact that he has various other beliefs, e.g. the belief that
there is some painful malady x such that x is referred to as “arthritis” and I have been told
that I have x. There is no possible world where Smith just believes S. A belief in S neces-
sarily presupposes some more fundamental belief, such as the one just described.
But in that case, Smith and Smith* have different beliefs in virtue of having the
very same beliefs. In virtue of having various existential beliefs, Smith has thoughts
about arthritis. In virtue of having the very same existential beliefs, Smith* has thoughts
about arthritis*. Moreover, Smith’s arthritis-thoughts are incompatible with Smith*’s
arthritis*-thoughts. Thus, if Burge is right, two people have incompatible beliefs in
virtue of having the very same beliefs.
To block this, Burge would have to say that it is not solely in virtue of having these
various other beliefs that Smith (Smith*) believes in S (S*). So even though Smith and
Smith* share existential-descriptive beliefs like there is some malady x such that x is
 Conceptual Atomism and the Computational Theory of Mind

called “arthritis” and such that x leads to excruciating joint-pain…their having these
other beliefs is not the sole reason why they believe in S and S*, respectively. There is,
Burge would have to maintain, an ineliminable causal component, it being that purely
causal, non-descriptive component that makes all the difference.
The problem is that this causal connection operates by way of knowledge that Smith
and Smith* have of the already discussed existence-claims. For example, suppose that,
having never before heard the term “arthritis” or any synonym, Smith’s doctor says to
him: “you are suffering from a disease known as ‘arthritis’, and that is why you are expe-
riencing pain and stiffness in your joints.” In virtue of hearing this, Smith is indeed
causally connected to an ailment (or, more accurately, to instances thereof) to which
Smith* is not connected; and, in virtue of having a corresponding experience in his
world, Smith* is causally connected to (instances of) an ailment to which Smith is not
connected. But whatever causal connection Smith has to arthritis that Smith* lacks,
that connection obviously operates by way of sense-perceptions of the sort just de-
scribed, and thus by way of Smith’s intake of purely existential information. It would
make no sense to say that, apart from such perceptions, Smith had any kind of relevant
causal connection to instances of arthritis. (Of course, exactly similar remarks apply to
Smith*.) So once we take into account their purely descriptive beliefs, we have already
fully taken into account any relevant causal connections to external objects that Smith
and Smith* might have. There is no causal connection over and above those involved in
the aforementioned descriptive beliefs to account for the supposed fact that Smith and
Smith* have different thoughts.

The debate between externalism and internalism not terminological

One might think that the debate between externalism and internalism is just a matter
of nomenclature. We can choose to let the word “perception” refer either to a certain
kind of purely psychological phenomenon or only to successful occurrences of that
phenomenon. (We can choose to refer to a drunkard’s hallucinations as “perceptions”
or we can choose to use that term to refer only to veridical perceptions.) Similarly, we
can choose to use the word “thought” to refer either to a purely psychological phenom-
enon or only to successful instances of that phenomenon. The externalist chooses the
one option; the internalist chooses the other.
This is not a tenable position. A veridical perception decomposes into a purely
psychological part and a non-psychological part. As we’ve seen, the non-psychological
part is what is described by the purely psychological part. That is why the purely psy-
chological part would have been a failure if the non-psychological part hadn’t existed.
Similarly, if thoughts decomposed into a psychological part and a non-psycholog-
ical part, then the psychological part of a thought would be something which would be
a failure if the non-psychological part didn’t exist, the reason being that the non-psy-
chological part would be what was described by the purely psychological part. So in
 Chapter 11.  Why one’s mental content is fixed by one’s epistemic situation 

order for thoughts to decompose into both a psychological part and a non-psycholog-
ical part, the purely psychological part must itself be a thought; it must itself be true or
false. But this means that the thought would be purely psychological, and that what we
were just describing as the non-psychological part of the thought wouldn’t be a part of
the thought at all, and would instead be what made the thought true.
 chapter 12

Jackson and Pettit on program-causality

and content-externalism

Content-externalism is incompatible with the presumption that, in virtue of having

representational properties, a brain-state has causal properties that it would not other-
wise have. Content-externalism is therefore incompatible with the fact that mental rep-
resentations do anything. In any case, this is what we argued in Chapter 1. John Mc-
Dowell tried to show otherwise. But we saw that McDowell’s argument is a
non-starter.
But there is an argument for that thesis that is far more powerful than any thus far
considered. That argument is due to Frank Jackson and Philip Pettit. Working jointly, those
authors have produced an original and compelling analysis of the concept of causal expla-
nation, and they have argued that, given this analysis, there is no difficulty reconciling
content-externalism with our presumptions concerning the causal potency of the mental.
I should begin by stating my position. I believe that what Jackson and Pettit say
about causal explanation in general is important and correct and cogently argued. But,
I will argue, the same cannot be said of their attempt to use these correct and deep
insights to rescue content-externalism. In their attempt to save content-externalism,
Jackson and Pettit have misapplied their own correct views.
Jackson and Pettit have also attempted to use their insights into causation to defend
a doctrine known as functionalism against what appears to be a devastating objection to
it. (Functionalism is a widely held view as to the nature of mental states. In a moment, we
will say more exactly what functionalism is.) We will find that, once again, Jackson and
Pettit have misapplied their own insights into causation, and we will also find that the
aforementioned criticism of functionalism is in fact cogent.

Why functionalism seems to be incompatible with the causal potency

of the mental

We have already seen why, supposedly, content-externalism is incompatible with the causal
potency of the mental. So let us discuss why the same appears to be true of functionalism
First of all, what is functionalism? Functionalism is the view that, for any instance of x
of any mental category M, it is wholly in virtue of what its causes and effects are that x
falls into M. Let m be the category belief that snow is white. (Thus, for x to be an instance
of m is for x to be a belief that snow is white.) Instances of m are likely to have certain
 Conceptual Atomism and the Computational Theory of Mind

causes and certain effects. For example, they are likely to be brought about by certain
kinds of sense-perceptions (e.g. perceptions of white snow) and they are likely, in their
turn, to bring about certain other mental states (e.g. beliefs that snow is not green) and
certain actions (e.g. utterances of “snow is white”). Functionalism is the view that for x
to be an instance of m just is for x to have causes and effects of the kind just described,
the same thing mutatis mutandis being true of any instance of any mental category.
Before we say what is wrong with functionalism, we must recall that it is states of
affairs, not objects, that make things happen. The rock didn’t break the window. What
did so was the rock’s having various properties, e.g. its having a certain mass and mov-
ing with a certain velocity. Similarly, it is not brain-state β that does any causal work
– it is brain-state β’s having such and such thermal (or morphological or electrical or
representational or syntactic…) properties that does so.
Let B be the brain-state that realizes Smith’s belief that 1+2=3. According to func-
tionalism, B realizes that belief entirely in virtue of what its causes and effects are. So B
qualifies as such a belief because circumstances at least approximately like the follow-
ing hold. Given that Smith already believes that 3 is an odd number, B’s existence leads
Smith to believe 1 and 2 add up to some odd number. Given that Smith already believes
that Jones has one car, and two motorcycles, and no other motorized vehicles, B’s exist-
ence lead Smith to believe that Jones has exactly three motorized vehicles. Given that
Smith already believes that three is the successor of two, and that adding one to a
number is the same as generating its successor, Smith (an English-speaker) says “three”
when asked “what is 1+2?”
Of course, given only what we have said about Smith, he needn’t have a belief that
1+2=3. Smith’s possession of the dispositions just described could be explained with-
out imputing such a belief to him. A closely related point is that Smith’s belief that
1+2=3 is not exhausted by his having just those three dispositions. But the idea behind
functionalism is that Smith’s belief is exhausted by those three dispositions plus a large
number of other dispositions of a similar kind.
According to functionalism, B’s being a belief that 1+2=3 is identical with (or, in
any case, is wholly realized by) its having a certain functional role. So B’s being such a
belief consists in (inter alia) its leading Smith, under the right circumstances, to form-
ing the belief that 1 and 2 add up to an odd number; it consists in its leading Smith to
forming the belief that Jones has three motorized vehicles.
Here is the problem. If B’s being such a belief consists in its leading to these states of
affairs, then B’s being such a belief cannot cause them to occur. But it is quite obvious
that, under the right circumstances, B’s being a belief that 1+2=3 is a cause of his form-
ing the belief that 1 and 2 add up to an odd number and that, under the right circum-
stances, it is a cause of Smith’s forming a belief that Jones has three motorized vehicles.
So functionalism strips Smith’s belief that 1+2=3 of any causal powers. Functionalism
doesn’t strip B of causal powers. But it does strip B’s being a belief that 1+2=3 of causal
powers. (For example, it strips B’s being such a belief of the ability to lead Smith to say
“three” when asked “what is 1+2?”) Since functionalism deprives B’s being such a belief
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

of the ability to bring about the things that it obviously does bring about, that doctrine
is clearly wrong.
Another illustration may help. You see a glass of water. Because you are violently
thirsty – so thirsty that you no longer have any concern for matters of decorum and social
propriety – you grab the glass and drink its contents. B is the brain-state of yours that real-
izes your being in a state of thirst. Surely B’s being (or realizing) a state of thirst is at least
one of the causal determinants of your grabbing the glass and drinking its contents.
But functionalism says that B’s realizing a state of thirst consists in its leading to
your grabbing the glass; it says that B’s realizing such a state is constituted by its leading
to such a state. Cause and effect must be distinct; no state of affairs causes itself or any
constituent of itself. So B’s realizing a state of thirst cannot possibly be constituted by or
consist in its leading to your reaching for the glass. Therefore functionalism is wrong,
given that B’s being a state of thirst obviously does (or at least could) cause you to reach
for the glass. In general, by identifying content with functional role, functionalism
strips content of the power to have any effect on functional role.

The Jackson-Pettit response: causal relevance versus causal efficacy

Jackson and Pettit (2004, 2004c) argue that, despite the criticisms just made, function-
alism is consistent with the presumption that the mental is causally potent. And they
have also argued that, despite the points made in Chapter 1, the same is true of con-
tent-externalism. Let us now consider what Jackson and Pettit say.
Consider the following statement:
(*) Pan P is hot at time t because, for the ten minutes preceding t, there was a
flame underneath P.
Let us suppose that, in world W, (*) describes an actual causal relationship, i.e. let us
suppose that (*) is as true a causal statement as a lay-person (someone who has not
mastered the subtleties of micro-physics) will ever produce. W is governed by the same
physical laws that govern our world. (W may even be our world.)
Granting all of this, it was not, strictly speaking, the flame that caused the pan to
heat up. It was not even the flame’s having a certain temperature. The only events that
really did any work in the way of heating up the pan were various mass-energy dis-
placements at, or vanishingly close to, the relevant surface of the pan. Let M1…Mn be
the mass-energy displacements at the point of contact between the flame and under-
side of the pan. Now consider a scenario W* that is just like W, with just one difference:
in W*, the flame in its entirety is absent except for M1…Mn. In W*, the pan will have
the very same temperature that it has in W. It seems, then, that the flame didn’t do any-
thing, except in so far as it comprised M1…Mn. This seems equivalent to saying that it
was M1…Mn, and not the flame, that heated the pan. If this is right, then (*) is false.
 Conceptual Atomism and the Computational Theory of Mind

Leaving aside the causal statements made by a few microphysicists, what we just
said about (*) is true mutatis mutandis of every causal statement that has ever been ut-
tered. Consider the statement:
(**) Because it is square shaped, the peg could not fit into the round hole.1
Suppose that (**) holds in world W, where W is governed by the same laws as ours. As
Jackson and Pettit point out, it was not the squareness of the peg that prevented it from
entering the hole; it was specific resistances generated by specific interactions of spe-
cific parts of specific surfaces. In order for those resistances to be generated, it isn’t
necessary that the peg be square. The corners of the peg are not, in this context, doing
anything: if we sawed them off, the peg still wouldn’t go into the hole (assuming, of
course, that we sawed off only modest amounts); we’d be left with a heptagon that
would fail to go into the hole for the very same reasons as the square. The forces that
prevent the peg from entering the hole are invariant with respect to infinitely many
different possible alterations of the peg’s shape and size, so long as those alterations fall
within certain limits. Thus, it seems false to say that the peg’s squareness is what pre-
vented it from entering the hole. Rather, what we must say is that specific forces (or
mass-energy displacements or…) F1…Fn, occurring at contact points C1…Cm, are what
prevented the peg from entering the hole.
But surely it wouldn’t be right say, without qualification, that (*) and (**) were
false. It is a datum that (*) and (**) are (or can be) true. Leaving aside the utterances of
a few microphysicists, not a single causal statement produced by anyone contains any
mention of anything comparable to M1…Mn. It is a datum that true and informative
causal statements are almost always given by statements like (*) and (**); and, notwith-
standing the points made a moment ago, it is therefore not an option to say, without
some kind of heavy qualification, that they are false.
One last example. The pressure of the gas inside the container is increased (because
that gas is heated). The container subsequently bursts. What caused the vessel to burst
was not, strictly speaking, the gas’s having (or suddenly acquiring) a certain pressure.
Ultimately it was specific impacts of specific particles on specific parts of the surface of the
container that led to the bursting of the vessel. Given a counterfactual scenario where
those specific impacts occurred, but where the container was otherwise empty of gas, the
container still would have burst. Given this, it seems that it is false to say that the gas (or
the gas’s having a certain pressure) is what caused the vessel to burst. The gas’s having that
pressure was completely innocuous, except in so far as it comprised the specific collisions

1. This example is borrowed from Jackson and Pettit (2004) who, in turn, borrowed it from
Putnam. We are, of course, making assumptions as to the comparative sizes of the peg and the hole,
and also as to the degrees of rigidity of the peg and of the material surrounding the hole.
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

just mentioned. At the same time, there is clearly some sense in which the gas’s pressure
is responsible for the bursting of the vessel. If I say:
(***) the container burst because the walls of the container couldn’t withstand the
pressure put on them by the gas inside,
my statement is not in the same category as “the container burst because elves used
their telepathic powers to make the container explode.” How is the putative truth of
(***) to be reconciled with the fact it was certain specific collisions K1…Kn, and not the
gas’s pressure per se, that ruptured the vessel? How are the truth of (*) and (**) to be
reconciled with the fact that, strictly speaking, it was E1…En and F1…Fn, and not the
flame or squareness of the peg per se, that heated the pan and prevented the peg from
entering the hole?
Jackson and Pettit deal with this problem by distinguishing between what they call
“causal efficacy” and what they call “causal relevance.” It must be kept in mind that
these are neologisms that are to be understood strictly in terms of the definitions that
Jackson and Pettit provide. Their pre-existing meanings are to be set aside (except in
so far as those coincide with the meanings given to them by Jackson and Pettit). In
connection with this, Jackson and Pettit distinguish between what they call “program
causes” and what I will henceforth refer to as “effective causes.” (The term “effective
cause” is my term. A state of affairs E is an “effective cause of ” of E* iff E is causally ef-
ficacious with respect to E*. Any similarities between my use of the term “effective
cause” and Aristotle’s are unintended and are to be disregarded.) Let us now define
these terms – or, in any case, let us illustrate their meanings.
Consider statement (*). Events E1…En are causally efficacious with respect to the
heating of the pan. They are causes of the heating of the pan in the strictest possible
sense. As we discussed, the flame itself is quite inert as regards the heating of the pan,
except in so far as it comprises E1…En. But, as we also pointed out, there is plainly a
sense in which (*) is true; there is plainly a sense of the word “cause” in which the flame
(or, more precisely, it’s having such and such a temperature) is a cause of the heating of
the pan. The flame (or its having such and such a temperature) is, as Jackson and Pettit
put it, “causally relevant” to the heating of pan. If one event is causally relevant, in the
relevant technical sense of that expression, to another, then the first “programs for” the
second or, equivalent, the first is a “program cause” of the second. So the flame is the
“program cause” of the heating of the pan. By contrast, E1…En are the “effective causes”
of the heating of the pan.
Let us consider statement (**). F1…Fn, occurring at contact points C1…Cm, are
what prevent the peg from entering the hole. But it is a datum that (**) is (or could be)
a true causal statement. If you ask me “why isn’t the peg going into the whole?” and I
say “because the peg is square, and the hole is round”, what I have said is not in the
same category as “because the square is green and green objects are physically incapa-
ble of entering apertures of any size.” The squareness of the peg is causally relevant to
its not being able to enter the hole; the former programs for, i.e. is the program cause,
 Conceptual Atomism and the Computational Theory of Mind

of the latter. F1…Fn, on the other hand, are the effective causes of the peg’s not entering
the hole, i.e. the former are causally efficacious with respect to the latter.
The gas’s exerting a pressure on the walls of the container is causally relevant to, i.e.
programs for, the rupturing the container; collisions K1…Kn are the effective causes of
the rupturing of the vessel, i.e. they are “causally efficacious” with respect to that event.

Precisifying the terms “program cause” and “effective cause”

These remarks help to make it clear at an intuitive level what is meant by the terms
“program cause” and “causally relevant.” Let us now try to convert that intuitive under-
standing into an explicit definition.
Although the flame per se is not causally efficacious with respect to the heating of
the pan, the existence of the flame guarantees the occurrence of events that are thus
efficacious. Let W^ be any counterfactual scenario where E1…En do not occur, but that
is otherwise just W (the world where (*) is true). Of course, in W^, the flame exists,
and the relation that it bears to the pan in that world is comparable to the relation that
it bears to the pan in W. Thanks to the existence of the flame, the pan will be heated in
W^: it is guaranteed that, in W^, some micro-events will have the effects on the pan
that, in W, are generated by E1…En. Under the circumstances, the flame’s presence thus
programs for the heating of the pan: it constitutes a structure that, under the circum-
stances, renders inevitable the heating of the pan. Equivalently, the flame’s presence is
causally relevant to the heating of the pan. (Remember that, in this context, the term
“causally relevant” is being used in a neologistic sense.)
Analogous remarks apply to each of the scenarios discussed. Although the square-
ness of the peg per se is not efficacious with respect to the peg’s resisting being put in
the hole, the squareness of the peg guarantees the presence of forces that are thus effi-
cacious. Given any world where F1…Fn (the forces that, in actuality, prevent the peg
from entering the hole) are absent, but is otherwise just like our world, the fact that the
peg is square guarantees that some forces will prevent the peg from entering the hole.
The squareness of the peg constitutes a structure that, under the circumstances, pre-
vents the peg from entering the hole, even though this outcome is not generated by the
squareness of the peg per se.
Let us turn to our third example. Under the circumstances described, it is inevita-
ble that the container will burst, given that the gas inside is exerting such and such a
degree of pressure on its walls. The gas’s having that pressure guarantees that, even if
K1…Kn specifically are absent, some collisions or other will cause the container to rup-
ture. Under the circumstances, the gas’s having that temperature programs for (or,
equivalently, is causally relevant to) that outcome.
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

How Jackson and Pettit apply the distinction between efficacy and relevance

Let us now discuss how, given the distinction between causal relevance and causal ef-
ficacy, Jackson and Pettit attempt to show that content-externalism is in fact compati-
ble with the presumed potency of the mental. Let W be our world, and let W* be a
world that is just like ours except that it contains XYZ as opposed to H2O. In W, I see a
glass of water (H2O) and, being thirsty, I decide to drink its contents. In W*, I see a
glass of XYZ and, being thirsty, I decide to drink its contents. The content-externalist
says that the contents of my perception and subsequent intention in W are different
from the contents of my perception and subsequent intention in W*. In W, I see, and
wish to drink, a glass of H2O. In W*, I see, and wish to drink a glass of XYZ.
Here is what the internalist says in response. What does all the work is what is
spatiotemporally local. What is remote does no work, except in so far as it leads to the
right local states of affairs. So what is in the glass is irrelevant to what I think or do,
except in so far as it leads to the right local events. Equivalently, given the right local
events, it is irrelevant, so far as the generation of my ideation and conduct are con-
cerned, what remote states of affairs obtain (or used to obtain). Thus, in so far as a
mental state’s content is a function of what is remote, its content is inert; a mental
state’s content is not inert only in so far as that content is a function of what is
local. Mental content does nothing, and thus has no effects, in so far as the content-
externalist’s conception of it is correct. Given that mental content does have effects, the
content-externalist’s position must be wrong.
Jackson and Pettit counter-respond by saying that this argument fails to distinguish
between causal relevance and causal efficacy. My thinking about H2O is not causally ef-
ficacious with respect to forming an intention to drink the contents of the glass. But given
that I am thinking about H2O, it is guaranteed, under the circumstances, that I will form
that intention. True – my thinking about H2O does not supervene on what is local (intra-
cranial). True – what is causally efficacious is confined to what is local. But it doesn’t fol-
low that my thinking about H2O isn’t causally relevant to what I think and do. Just as the
presence of the flame beneath the pan guarantees that the pan will be heated, so my
thinking about H2O guarantees that I will generate an intention to drink from the glass.
The relationship between my forming an intention to drink H2O and my subse-
quent behavior is perfectly comparable, according to Jackson and Pettit, to the rela-
tionship between the presence of the flame beneath the pan and the subsequent heat-
ing of the pan. Outside the arena of microphysics, the latter is as robust an instance of
a causal relationship as we will ever find. The same is therefore true of the first of the
former relationship, given that the two relationships are comparable. So content-exter-
nalism by no means belittles the causal potency of the mental.
Remember our story in Chapter 1 about Max and Twin-Max. Here is what Jackson
and Pettit would say about that situation. Given that Max is thinking about R, it is
guaranteed that he will think and act in certain ways, just as, given the squareness of
the peg, it is guaranteed that the peg will not enter the hole (even though the square-
 Conceptual Atomism and the Computational Theory of Mind

ness of the peg is not causally efficacious with respect to the generation of that out-
come). R’s being a constituent of Max’s mental content is therefore causally relevant to
what Max thinks and does. Thus, content-externalism allows mental content to be as
causally potent as the squareness of the peg or the temperature of the flame.

Evaluating the Jackson-Pettit view

First of all, the distinction between causal relevance and causally efficacy – between
program causes and effective causes – is an entirely correct one; and it is, in my view,
of the highest importance.2 Explanation almost always consists in identifying pro-
gram-causes, not effective-causes; and this is not simply because we are not always in
a position to identify effective-causes. Jackson and Pettit convincingly argue that one
is typically given much more predictive, explanatory, and counterfactual information
if one finds out what the program-causes of an event are than if one learns what its ef-
fective-causes are (Jackson and Pettit 2004d). Their arguments on this matter strike me
as cogent, and I have nothing to add to them.
But we must regard as fallacious their attempt to show that, given the idea of a
program-cause, content-externalism is consistent with the presumed causal potency of
the mental. There are several reasons for this.
Suppose we want to know whether Max’s thinking about R has any causal powers
at all – whether it is either causally efficacious or causally relevant to anything Max
thinks or does. Fodor has made it clear that the relevant question is not “given that
Max is thinking about R, is there some kind of guarantee that he will do such and
such?” Rather, the relevant question is: “given that Max is thinking about R, is there
some kind of guarantee that he will do such and such that would be absent if he were
thinking about R*?” More simply: “Does Max’s thinking about R have any causal prop-
erties that his thinking about R* would not?” The externalist says: “yes – because he is
thinking about R, not R*, he has a desire to take a photograph of R, but of R*; he has a
surge of joy in response to R’s aesthetic properties, and not in response to R*’s.”
But that answer, we have seen, is irrelevant. Remember what we said in Chapter 1.
I quote:
Magnet A is attracted to magnet B. Magnet B happens to be green. But (ceteris
paribus) if B were red or purple or orange, A would still be attracted to it. It is true
that, because B is green, A is behaving in a way that it would not behave if B were
some other color. (A is drawn towards a green magnet, instead of a red one.) But
it doesn’t follow that B’s color has anything to do with A’s behavior. A’s behavior
wouldn’t (ceteris paribus) be relevantly different in a world where B were some

2. In my doctoral dissertation, The Moral Structure of Legal Obligation, I argued that that the
Jackson-Pettit analysis of causality makes it possible to solve a number of otherwise unsolvable
puzzles relating to the concepts of institutional behavior and legal obligation.
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

other color. This shows that B’s being green has no effect on what A does, notwith-
standing that, because B is green, A is doing something (namely moving towards
a green magnet) that it wouldn’t be doing if B were red.

For exactly similar reasons, given only that Max has a desire to take a picture of R, as
opposed to R*, it doesn’t follow that his thinking about R has any effect on his thought
or conduct that this thinking about R* would not. The distinction between causal effi-
cacy and causal relevance is irrelevant here. The argument just given shows that Max and
Twin-Max don’t differ in respect of any of their causal properties. If Max were to trade
places with Twin-Max, the one would program-cause exactly what would otherwise be
program-caused by the other. Therefore Max and Twin-Max don’t differ in respect of
their program-causal properties. This shows that, so far as mental content is to be de-
scribed along content-externalist lines, differences in content are program-causally in-
nocuous and, therefore, that mental content itself is program-causally innocuous.
The Jackson-Pettit argument involves a failure to distinguish between the logical
and the causal dimensions of explanation. Ten minutes ago, the pan was cold. Now it
is hot. I want an explanation. (To facilitate exposition, let’s suppose that I haven’t yet
learned that fire heats.) I consider the various events that preceded the heating of the
pan. I put on a green shirt (I was previously wearing a blue one). I played the piano. I
turned on the light in the kitchen. I turned on the burner on which the pan was resting.
I brushed my teeth. To figure out which of these events was responsible for the heating
of the pan, I take a tour of all those worlds that are just like this one, except that each
fails to comprise exactly one of the events just mentioned. (The epistemological equiv-
alent of this tour could be accomplished through experimentation – through employ-
ment of Mill’s methods, or some such. We can regard talk of this metaphysically im-
possible tour as a way of condensing a long and tedious description of experiments
that could be conducted within our world.) So in W1 I don’t put on a blue shirt; but
otherwise W1 is just like our world. In W2 I don’t brush my teeth; but otherwise W2 is
just like our world. And so on. Of course, I notice that in all but one of these worlds the
pan is hot. The one where the pan is cold is the one that is just like ours except that, in
it, I don’t turn on the burner. Let W7 be that world.
In W1, the pan does indeed have a property that it doesn’t have in our world – it has
the property of being such that it is within ten feet of somebody who is not wearing a
blue shirt. And in W2, the pan does indeed have a property that it doesn’t have in our
world – it has the property of being such that it is within ten feet of somebody who
hasn’t recently brushed his teeth. But this obviously isn’t enough to show that, in our
world, my wearing a blue shirt or brushing my teeth is a program-cause of the pan’s
being heated. Leaving aside the fact that, in W2, I haven’t brushed my teeth, the pan’s
behavior there is identical with the pan’s behavior here. This point is easily generalized.
Let e1…en be the events described (so e1 is the event of my brushing my teeth; e2 is the
event of my putting on a blue shirt; and so on); and let Wi be a world that is just like ours
except that it doesn’t comprise ei. For each i, other than i=7, Wi is just like our world
 Conceptual Atomism and the Computational Theory of Mind

leaving aside the fact that Wi doesn’t comprise ei. This shows that e7 (the event of my turn-
ing on the burner) is the culprit. e7 is what is responsible for the pan’s heating.
Of course, e7 is a program-cause of the pan’s heating, i.e. the former is causally rele-
vant to the latter, but not causally efficacious with respect to it. Given only that W7 is just
like our world, except that it doesn’t include the event of my turning on the burner, it
logically follows that W7 has properties not had by our world. For example, it logically
follows that the pan has the property of not being such that, at a certain time, it is within
a certain distance of someone who has turned on a certain kitchen appliance; and it
logically follows that the pan has the property of being such that it is either a square circle
or JM didn’t execute any intention he might have had to turn on a certain burner at a
certain time. The reason that e7 is a program-cause of the pan’s heating is that leaving
aside properties of the kind just mentioned, W7 differs from our world.
Causal relations are given by synthetic propositions. Logical relations are not
causal relations – they are not relations of effective causality or of program-causality. A
corollary is that, in so far as statement P provides a logical guarantee of statement Q’s
truth, P doesn’t provide a causal guarantee of Q’s truth. The statement at t, JM turned
on the burner doesn’t logically guarantee the truth of the statement the pan heated up.
That is why the first statement can, at least in principle, provide a causal explanation
for the truth of the second. The statement at t, JM did something which would eventuate
in the pan’s heating up logically guarantees the truth of the pan heated up, and thus
provides no causal explanation for that truth.
To identify the cause (in any sense of the word) of an event, one considers situa-
tions that differ in some specific respect from the one that led to that event, but that are
otherwise like the latter. But one must set aside the strictly logical consequences of the
differences just discussed. Given that, in W3, I am wearing a green shirt instead of a
blue one, it logically follows that the pan is such that it is within the vicinity of a person
wearing a green shirt. So given our description of that world, it is indeed guaranteed
that the pan has that property. But that guarantee is of an analytic, not a causal, nature.
By contrast, given that, in W7, I don’t turn on the burner, nothing concerning the pan’s
temperature is logically guaranteed; and any outcome as to the pan’s thermal state is
given by a synthetic, and therefore at least potentially causal, statement.
Jackson and Pettit are conflating the concept of a logical guarantee with that of a
causal guarantee. Given how we’ve described Max’s and Twin-Max’s respective worlds,
it is indeed guaranteed that Max will behave differently from Twin-Max. But that guar-
antee is logical: it is an artifact of the content-externalist’s description. It isn’t compara-
ble to the guarantee provided by JM turned the burner on of the truth of the pan subse-
quently heated.
Let us close the argument. As before, let W be Max’s world – the world where he
sees rock R. If it is supposed that there is a world W* that is just like W, except that W*
doesn’t comprise R, it logically follows that Max will have many properties that he
doesn’t have in W. For example, in W*, Max will have the property of not being within
a hundred feet of R at time t. If the content-externalist is right, another one of these
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

properties is that Max will have the property of not being such that R is a constituent of
his mental content. To establish that Max’s thinking about R is a program-cause of any-
thing that he thinks or does, we need to show that leaving aside these differences – i.e.
leaving aside the differences that follow analytically from our descriptions of W and W*
– Max’s thought and behavior in W* is different from his thought and behavior in W.
But this is precisely what we find not to be the case. Leaving aside the (supposed)
fact that, in W*, Max does not have R as a constituent of his thought; and leaving aside
other differences between W and W* that follow analytically from our description of
those two worlds; there are no differences between Max as he is in W and Max as he is
in W*. So if R is a constituent of Max’s mental content, that fact is no more causally
relevant to his thought or conduct than my putting on a blue shirt is causally relevant
to the heating of the pan.
Another example may be appropriate. Given that I am seeing a dewy glass of ice-
water, I will indeed do something that I wouldn’t do if (ceteris paribus) I were seeing a
dewy glass of XYZ. Because I am seeing H2O, I will reach out and grab a glass of H2O.
If (ceteris paribus) I were seeing XYZ, I would reach out and grab a glass of XYZ. But
this doesn’t show that my seeing H2O is causally relevant to my thought or conduct.
Given only our description of the XYZ-world, it immediately follows that I have many
properties there that I don’t have here. For example, it follows that, in that world, I
consist mostly of XYZ, as opposed to H2O. (That logically follows given our pre-exist-
ing knowledge that people consist mostly of H2O.) If the externalist is right, it also
immediately follows that I have thoughts about XYZ in that world, but not in this one.
But to show that my thinking about H2O is causally relevant to my behavior or thought,
we need to show that leaving aside those analytic differences my thought and conduct
in the XYZ-world are different from my thought and conduct in this world.
That condition is obviously not met. Jackson and Pettit contend that my having
thoughts about H2O bears the same relation to my subsequently grabbing the glass that
the presence of the flame bears to the subsequent heating of the pan. This contention
is false. Given a world just like ours, except that it doesn’t comprise the event of my
turning on the burner, the pan will not heat: such a world will be different from ours
even after we set aside the differences that logically follow from our description of it. That
is why there is, or at least could be, a causal relation (albeit one of relevance, not effi-
cacy) between the flame’s presence underneath the pan and the pan’s temperature. By
contrast, given a world like ours, but where there is XYZ in the glass, I still reach out
and grab the glass: such a world is not different from ours after we leave aside the dif-
ferences that logically follow from our description of it.
Let us suppose for the sake of argument that the content-externalist is right. In
that case, the glass is a veritable component of my mental content. Given that the glass
is such a component, it is indeed guaranteed that some events will occur that are caus-
ally efficacious with respect to my forming the previously described intention. But
here we are dealing with a different kind of guarantee from those previously discussed.
As the content-externalist himself insists, the glass’s being a component of my mental
 Conceptual Atomism and the Computational Theory of Mind

states’ content does not supervene on what is local; it supervenes on what is remote –
on what is, in fact, non-existent by the time that B1…Bn occur (or by the time that there
occurs anything else that might be causally efficacious with respect to my forming an
intention to grab the glass). The glass’s being a constituent of my mental states’ con-
tents therefore doesn’t comprise anything that could possibly be an effective cause of
my forming that intention or, therefore, of the behavior that expresses that intention.
Therefore the glass’s being such a constituent doesn’t guarantee the formation of such
an intention in a sense comparable to that in which the presence of the flame guaran-
tees the heating of the pan. In the one case, the guarantee in question is an artifact of
our description of the situation; in the other case, the guarantee holds in virtue of a
causal nexus that has nothing to do with our descriptions.
For the content-externalist, the glass’s being a constituent of my mental states’ con-
tents consists in the glass’s having certain effects on me – it consists in the fact that
some state of affairs involving the glass is the (distal) cause of certain changes in my
person. Content-externalism just is the position object X’s being a constituent of the
content of mental state Y consists in its being the case that X (or, more accurately, some
state of affairs involving X) causes Y.
So given an acceptance of content-externalism, it follow tautologously that, if X is
a component of mental state Y, then X (or some state of affairs involving X) is what
caused Y to occur. In particular, it follows tautologously that, if the glass is a constitu-
ent of the content of some mental state of mine, the glass is a cause of that state.
Thus, if we accept content-externalism, then the statement “the glass is a compo-
nent of the content of mental state M had by JM” tautologously implies the statement
“N was caused by the glass (or, more exactly, by some state of affairs involving the
glass).” Given that any neural state of mine will be causally potent, it follows that the
statement “the glass is a component of the content of neural state M had by JM” guar-
antees the truth of some statement ascribing a causally potent state to me.
But this guarantee follows logically from the content-externalist’s position; it is a
by-product of his way of describing the situation. And this is exactly the kind of guar-
antee that we don’t want in this context. For as we discussed earlier, when we want to
assess whether the glass’s being a constituent of my mental states’ content has any kind
of causal potency, the question to ask is this:
Consider a world W^ where that particular glass is absent, and where some other
glass has taken its place, but that is otherwise just like our world. Leaving aside the
differences between W^ and our world that follow logically from our description
of W^, is there any difference between how I think and act in W^ and how I think
and act in this world?

The answer to that question is obviously “no.” By hypothesis, the proximal causes in
W^ of my thought and behavior would be identical with their proximal causes in our
world. True – in W^, those proximal causes would lead to my grabbing glass X as op-
posed to glass Y. But given how we’ve described W^, it follows tautologously that it is
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

X that I will reach for, and not Y. When we wish to assess causal potency – whether in
the form of causal efficacy or causal relevance – we obviously don’t want to study the
logical consequences of our definitions. Supposing that the content-externalist’s posi-
tion is correct, it is indeed guaranteed that the glass’s being a constituent of my mental
content has various effects on what I think and do. But the guarantee in question is of
a logical, indeed a tautologous, nature; it is not a causal guarantee.
Another example may be appropriate. I am gazing at the heavens with my high-
powered telescope. I see star X. Because X is so beautiful, I am filled with joy. Given
this, let us suppose for argument’s sake that content-externalism is correct. In that
case, X is a constituent of the representational content of my perceptual and cognitive
states (in particular, of the aesthetic attitude just described). But, we may suppose, X
ceased to exist millions of years ago. (I can see it because it took millions of years for
the relevant light-rays to travel from X to my retinas.)
The externalist says that X is a constituent of the content of my visual perception.
Given that X doesn’t exist, it follows that, for the content-externalist, my perception does
not have a complete content: it has a gap in it where there is supposed to be a star. But my
perception does have a complete content, and content-externalism is therefore wrong.
It isn’t clear how the externalist is to respond to this, given that my visual percep-
tion has a content – a complete content, not one that has a gap in it where X is supposed
to be.3 But let us suppose for argument’s sake that content-externalism can deal with
this problem. Supposing that content-externalism is right, my thinking about X – i.e.
X’s being a constituent of the contents of my mental states – supervenes on processes
that went out of existence well before there were human beings. This makes it hard to
see how X’s being a constituent of my mental-content could have any causal potency.
After all, given the content-externalist’s position, X’s being a content of my thought
supervenes on what is buried in the remote recesses of space-time.
Here is what Jackson and Pettit (2004) say about this.
X’s being a constituent of my mental content programs for, i.e. is causally relevant to,
what I think and do. Admittedly, X’s being such a constituent isn’t an effective cause
of my thoughts or deeds. But the flame isn’t an effective cause of the heating of the
pan. So our position hardly downgrades the causal potency of representational con-
tent, given that it assigns as much causal potency to our thoughts as it does to flames.

3. Conceivably, the content-externalist could respond by advocating a four-dimensionalist

conception of existence: objects that have “ceased to exist” still exist, albeit in another temporal
place. But it seems suspicious that content-externalism – a doctrine about mental content – pre-
supposes the truth of such a view, even if that view happens to be correct.
Also, even if four-dimensionalism is correct, that doesn’t help the content-externalist. For it is
a fact that right now – at this time, i.e. in this temporal place – my visual perception has a complete
content; and, for the reasons just stated, content-externalism cannot accommodate that fact.
 Conceptual Atomism and the Computational Theory of Mind

The flame is contemporaneous with events that are causally effective with respect to
the heating of the pan. Indeed, the latter are constituents of the former, and therein lies
the nature of the causal relation between the flame and the heating of the pan.
But X ceased to exist millions of years ago. The same is true, of course, of the spe-
cific state of affairs involving X that is responsible for my perception (namely, X’s de-
flecting such and such light-rays in the direction of Earth, or some such). So nothing
that is an effective cause of anything that I think or do could possibly be a constituent
of any state of affairs involving X. Given this, it should be apparent that Jackson and
Pettit are over-extending the term “program-cause” – that they are using it to denote
two quite different concepts.
Supposing that content-externalism is correct, X’s being a constituent of my men-
tal content does indeed guarantee that there will be certain modifications of my neural
(or mental) condition. But the nature of this guarantee is logical, not causal, and it is a
tautologous consequence of the content-externalist’s position. That position is the view
that X’s being a constituent of my mental content is identical with X’s being a cause (of
some kind) of some current, and therefore causally potent, condition of mine. So given
the content-externalist’s position:
X is a component of the content of JM’s mental content =DF X (or, more accurately,
some state of affairs involving X) caused some current, and therefore causally po-
tent, modification of JM’s mind or nervous-system to occur.

By contrast, the connection between the presence of the flame and the heating of the
pan – or between the temperature (pressure) of the gas and the rupturing of the vessel
or between the squareness of the peg and its resistance to entering the hole – is en-
tirely synthetic and a posteriori. That connection is given by a synthetic a posteriori
proposition, and is therefore not a tautologous consequence of our definitions.

What led Jackson-Pettit astray

As we’ve seen, the Jackson-Pettit analysis involves a failure to distinguish between logical
and causal explanations (or, rather, between the logical and causal aspects of explanation).
This is related, I now propose to show, to their not sufficiently distinguishing between:
(a) X is a cause of so and so’s current condition
and:
(b) X’s being a constituent of so and so’s mental state is a cause of so and so’s cur-
rent condition.
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

There is an important difference between saying:

(1) Alpha Centauri (or some state of affairs involving that star) is a cause of Smith’s
current thought and behavior.
and saying:
(2) Alpha Centauri’s being a constituent of Smith’s mental content is a cause of
Smith’s current thought and behavior.
Here is an example of what I mean. Smith and Jones are both content-externalists.
Smith is looking at star X. Because X is so beautiful, Smith experiences an aesthetic
epiphany and thus decides to compose a poem celebrating X. Smith and Jones both
know that X went out of existence long ago. Smith asks Jones: “How is X’s being a con-
stituent of my mental content causally potent with respect to anything that I now think
or do, given that X ceased to exist long ago and that everything that governs my con-
duct and mentation is in the here and now?” Having read Jackson and Pettit, Jones
answers by saying:
(E) You are seeing X because some state of affairs involving X caused you to have
the visual experience that you are now having. Your having that visual experi-
ence is what caused you to have the epiphany that you are now experiencing.
So (given that your visual experience efficiently caused, and therefore causally
guaranteed, your subsequent epiphany) your having that epiphany was guar-
anteed to occur by the fact that you saw X and, therefore, by the fact that your
visual experience had certain origins. In other words, your having that epiph-
any was programmed for by your seeing X and, therefore, by your being in a
state whose distal cause was some state of affairs involving X.
Jones is obviously right that X (or some state of affairs involving X) is a cause of Smith’s
current neural condition and thus of his current mentation. But it doesn’t follow that
X’s being a constituent of Smith’s mental content is a cause of anything having to do with
Smith’s current thought or behavior. Given the content-externalist’s position, if X is to
be a constituent of Smith’s mental content, it is analytically necessary that some state of
affairs involving X be a cause of some aspect of Smith’s current condition. So given the
content-externalist’s position, the proposition:
(A) X is a constituent of Smith’s mental content
entails:
(B) X (or, more accurately, some state of affairs involving X) is a cause of (some
aspect of) Smith’s current condition.
So given the content-externalist’s position, and given that X is a constituent of Smith’s
mental content, it is indeed guaranteed that some fact about X is causally responsible for
what X now thinks and does. But what obviously doesn’t follow from (A) and (B) is:
 Conceptual Atomism and the Computational Theory of Mind

(C) X’s being a constituent of Smith’s mental content is causally responsible for
what Smith is now thinking and doing.
What obviously is causally responsible for Smith’s current condition is the fact that, mil-
lions of years ago, X deflected some light-ray in Earth’s direction, and that light-ray even-
tually collided with Smith’s retinas. So, quite obviously, some state of affairs involving X
affects what Smith is now thinking and doing. But it doesn’t follow that X’s being a con-
stituent of Smith’s mental content is one of those states of affairs. In fact, it makes little
sense to speak of a “state of affairs” here at all.4 There is no time during which both X and
Smith exist. So it is hard to see how there could be any state of affairs, in any causally sig-
nificant sense of the term, consisting of X’s being a constituent of Smith’s mental content.
Consider the flame. There are infinitely many times t such that, at t, we can truly
say: the flame now exists. That is why, for many times t, the flame (or, in any case, its
being hot) is operational: it heats; it burns; it soothes; it torments. The same is true of
the squareness of the peg, the temperature of the gas, and anything else that is a rela-
tum of a true causal statement. There is no time t such that, at t, we can say: star X’s
being a constituent of Smith’s mental content is doing such and such. After all, there is no
time at which X’s being such a constituent exists, given that X and Smith are separated
by an interval of millions of years. Of course, if the externalist is right, the noun phrase
“X’s being a constituent of Smith’s mental content” is not an empty one. But not every
non-empty noun-phrase denotes a state of affairs, if by “state of affairs” we mean some-
thing of causal significance – something like the flame or the squareness of the peg.
After all “the fact that there are infinitely many primes” is not an empty noun-phrase
– it isn’t like “the man on the moon” – but it doesn’t denote anything that has effects.
Jackson and Pettit were misled by their own examples. The examples that we have
considered – the flame, the peg, the gas’s temperature – are among the examples used
by Jackson and Pettit (or, in any case, are only trivially different from their own exam-
ples). But Jackson and Pettit give additional examples of what they describe as “pro-
gram causes” (or as instances of “causal relevance”), and these other examples are not
instances of the concept that we have been discussing.
Here is one of the examples in question. During a performance, the symphony-
conductor becomes annoyed. Why? Because somebody coughed. Here is what Jackson
and Pettit (2004: 55–59) say. Obviously the proposition (or statement):
(SC) somebody coughed
provides a perfectly good causal explanation of the fact that the conductor became an-
noyed. But (SC) doesn’t identify the effective cause of the conductor’s becoming an-

4. At least not if the term “state of affairs” is to be taken in a causally significant sense. We could
conceivably speak of mathematical facts as “states of affairs.” Referring to the fact that there are
infinitely many primes, a mathematician might say: “given this state of affairs, it follows that Rie-
mann’s efforts were not totally misguided.” But that is obviously not a sense of “state of affairs” that
is relevant here.
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

noyed. After all, it cannot just be somebody who coughed: it must be Smith, Jones, or
Brown – it must be some specific person. There is no state of affairs – no causally sig-
nificant entity – consisting in somebody’s coughing. If (SC) is true, that is parasitic on
the truth of some singular proposition. But (SC) is still a correct causal explanation.
Why? Because, given the truth of (SC), it follows that, for some x, the singular proposi-
tion x coughed is true: and x coughed does identify a state of affairs that led to the rip-
ples in the quantum that are constitutive of the conductor’s becoming annoyed.
What is most significant in this context is what Jackson and Pettit don’t say about
this example. Given the truth of (SC), it does indeed follow that, for some x, the singu-
lar proposition:
(XC) x coughed
is true. x’s coughing explains the conductor’s becoming annoyed in approximately (but
only approximately: see below) the same way that the flame’s presence explains the
pan’s becoming heated. x’s coughing is not causally effective with respect to the con-
ductor’s becoming annoyed. What is thus effective are various neural events N1…Nn
that are confined to the interior of the conductor’s cranium. x’s coughing (or, in actual-
ity, some causal successor of x’s coughing) programs for the occurrence of N1…Nn.
In fact, what is going on is much more nuanced than Jackson and Pettit allow. First
of all, it isn’t every aspect of x’s coughing that is responsible, on any delineation of the
term “responsible”, for the conductor’s becoming annoyed. Rather, what happens is this.
Among the mass-energy displacements constitutive of x’s coughing are various micro-
events. Those micro-events lead to other micro-events. And so on, until we have some
set of micro-events that lead to various micro-events that are constitutive of some physi-
ological condition on the part of the conductor. That condition is related to the conduc-
tor’s becoming annoyed in the way that the presence of the flame is related to the heating
of the pan. Among the constituents of that condition are various micro-events that ef-
fectively-cause the micro-events constitutive of the conductor’s becoming annoyed.
Jackson and Pettit say that (SC) is related to the conductor’s becoming annoyed in
the same way that (*) is related to the pan’s becoming hot. In other words, they say that
somebody’s coughing (not this or that person’s coughing, but just somebody’s doing so)
is related to the conductor’s becoming annoyed in the way that the presence of the
flame is related to the heating of the pan. But that is false for many reasons. First of all,
unlike the presence of the flame, somebody’s coughing (as opposed to Brown’s cough-
ing or Smith’s coughing) isn’t a state of affairs. As Russell (1903) said, you never just
meet a person. You meet Bob or Mary or Jane. Similarly, you never just hear someone
coughing; you hear some specific person doing so (even though you may have no idea
who that person is).
It is true that (SC) does provide a perfectly good explanation as to why the conduc-
tor became annoyed. But the reasons for this are quite delicate and cannot be straight-
forwardly modeled on the relationship between (for example) the presence of the
flame and the heating of the pan. Given that (SC) is true, it follows logically that (XC)
 Conceptual Atomism and the Computational Theory of Mind

is true (for some x). Given that (XC) is true, and given various causal laws, it follows
that some state of affairs C holds, where C – not the non-entity described by (SC) –
programs for N1…Nn.
As before, Jackson and Pettit are conflating the logical and causal dimensions of
explanation. Given (SC), along with some principles of physics and psychology, it is
guaranteed (or made highly probable) that the conductor will become annoyed. But
not every guarantee is comparable to the guarantee that the flame’s presence provides
for the pan’s becoming hot. The guarantee provided by (SC) combines logical, effec-
tive-causal, and program-causal components. Given (SC), it logically follows that (XC)
is true (for some value of x). The state of affairs described by (XC) effectively causes
physiological condition C. C programs for N1…Nn, i.e. C is related to N1…Nn in a way
that parallels the relationship of the flame to the heating of the pan. And, finally, N1…
Nn effectively-cause the conductor to become annoyed. (N1…Nn effectively cause the
brain-states that, supposing materialism to be correct, realize or are identical with the
conductor’s being annoyed.) But somebody’s coughing isn’t related to the conductor’s
becoming annoyed in remotely the same way that the presence of the flame is related
to the heating of the pan. The flame is a state of affairs: it is right there, underneath the
pan. Somebody’s coughing is not a state of affairs, and thus cannot be related to anything
(e.g. the conductor’s becoming annoyed) in the way in which the flame is related to the
heating of the pan. So given that the flame’s presence underneath the pan is a para-
digm-case of a program-cause, Jackson and Pettit are hyper-extending the term “pro-
gram-cause” in saying that it applies to somebody’s coughing.
Jackson and Pettit’s final example further over-extends the term “program-cause”.
5
Particles A and A* are qualitatively identical, and are in qualitatively identical circum-
stances (at least as far as their immediate vicinities are concerned). But they are thou-
sands of miles from each other. At time t, particle B hits particle A, and particle B* hits
particle A*. B and B* are qualitatively identical, and they hit A and A*, respectively, with
the same degree of force. The result is that A is displaced by the same amount as A*.
Smith asks Brown why the following proposition is true:
(AM) A and A* move by the same amount.
Brown answers by saying that:
(SAF) A and A* were struck with the same amount of force.
Brown’s answer is obviously correct, and it provides a perfectly good causal explana-
tion of the fact that Smith’s question concerns. But the cause of a particle’s being dis-
placed must be vanishingly close to that particle. The expression “the fact that A and
A* are struck with the same degree of force” doesn’t refer to anything that is vanish-
ingly close to either A or A*. In fact, it doesn’t refer to anything that has any spatiotem-
poral location. So how can Brown’s answer be correct?

5. See Jackson and Pettit (2004: 55-59).

 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

Here is what Jackson and Pettit say. Given the truth of (SAF), along with some ac-
cepted principles of physics, the truth of (AM) is guaranteed. Of course, Jackson and
Pettit are right: (AM)’s truth is guaranteed. But the nature of that guarantee isn’t com-
parable to the guarantee that the flame’s presence provides for the heating of the pan.
As Jackson and Pettit themselves point out, the fact that A and A* are struck with the
same force isn’t a state of affairs. (The result of putting quotation marks around the
italicized expression doesn’t pick out anything that has spatiotemporal boundaries,
even of the most nebulous sort.) Brown is providing a good causal explanation. But
that explanation is more akin to that provided by (SC) than to that provided by (*).
Brown’s explanation combines logical and effective-causal components.
Let us start with (SAF). As a matter of logic, this entails that for some specific
value of F, some singular proposition of the form:
(AF) A is struck with F
is true, and that some singular proposition of the form:
(A*F) A* is struck with F
is also true.
Of course, each of (AF) and (A*F) identifies some state of affairs (in the relevant
sense). Given the laws of physics, A’s being struck with F is an effective cause of A’s
moving by amount Q. And given those same laws, A*’s being struck with F is an effec-
tive cause of A*’s moving by amount Q. So given the truth of both (AF) and (A*F), the
following proposition follows:
(BAF) both A and A* move by amount Q
And given (BAF), (SAF) logically follows.
But what is important here is that there is no state of affairs consisting in the fact
that A and A* are struck with the same force, and there is no state of affairs consisting
in the fact that A and A* are displaced by the same amount.
By contrast, the flame’s being beneath the pan at t and the pan’s having temperature T
at t* are both states of affairs: they are spatiotemporally located entities; the expressions
that denote them are not nominalizations of abstract relationships or generalizations
over singular propositions. Thus, the relationship between (SAF) and (AM) is not com-
parable to that holding between the flame’s presence and the pan’s temperature.6

6. One last example will help identify the fallacy in the Jackson-Pettit analysis. Pans P and P*
are qualitatively identical, and are under qualitatively identical circumstances, but are thousands
of miles apart. Flames of equal heat and volume are placed underneath them. They heat up by
the same amount. Smith asks Brown why the proposition:
(PH) P and P* heat up by the same amount
is true. Brown answers by saying:
(PA) Because P and P* were heated by the same amount
 Conceptual Atomism and the Computational Theory of Mind

Brown has provided a correct causal explanation of (PH). But how can this be, given that causes
operate locally? Jackson and Pettit would say the following. Given (PH), it is guaranteed that
(PA) is true; therefore (PH) gives a program-explanation of (PA)’s truth.
But given obvious extensions of what we just said, the relationship between (PH) and (PA)
isn’t comparable to the relationship between the peg is square and the peg won’t go into the hole.
In the latter case, we have a state of affairs (the peg’s being square) which consists in various
mass-energy displacements, which effectively cause some other state of affairs (the peg’s remai-
ning in a certain place above the template with the hole in it, despite my pushing against the peg
with all my might). We don’t have anything like this in the first case. After all, there is no state of
affairs consisting in flames of equal intensities being put under the pans. Of course, the expres-
sion “the fact that flames of equal intensities were put under the pans” is not an empty one – it is
not like “the king of France.” But that expression doesn’t denote a constituent of the spatiotem-
poral world; that expression thus isn’t like “the flame” or “the peg’s being square.”
The real relationship between (PH) and (PA) is this. (PH) entails that, for some specific
values of p1…pn, both of the following singular propositions are true:
(F) A flame with properties p1…pn was placed beneath P at t.
(F*) A flame with properties p1…pn was placed beneath P* at t.
Notice that (F) describes a state of affairs (or event). Further, (F) describes what, given Jackson and
Pettit’s own statements, must be described as a program-cause of P’s heating up by a certain amount.
What is involved in this last fact? Among the constituents of the flame that is placed underneath
P* are various events that effectively cause pan P to heat up by amount Q (by time t*).
For exactly similar reasons, (F*) gives a program-cause of pan P*’s heating up by amount Q
(by time t*); and this is because (F*) identifies a state of affairs (or event) among whose consti-
tuents are effective-causes of P*’s heating up by Q.
Thus, given accepted principles of physics, (F) renders probable
(FH) P heats up by amount Q by time t*,
and given those same principles (F*) renders probable some singular proposition of the form:
(FH*) P* heats up by amount Q by time t*.
Given that (FH) and (FH*) are both true, (PA) logically follows.
Here two points deserve special mention. One is a point we have already emphasized, na-
mely: neither (PA) nor (PH) describes a state of affairs. For reasons already discussed, that by
itself makes it questionable whether we are dealing with anything comparable to the relationship
between the squareness of the peg and the peg’s resistance to going into the hole.
The other point is one we have not yet made. By Jackson and Pettit’s own lights – by their
own explicit pronouncements – the fact that a flame with properties p1…pn is being placed
beneath P programs for P’s heating up by a certain amount (after a certain interval). At the same
time, Jackson and Pettit would also say that (PA) provides a program-explanation for (PH).
Whether this constitutes vicious circularity is a question I will leave aside. But it seems clear that
the relationship between (PA) and (PH) is of a different logical-order from the relationship
between P being put on a flame with certain properties and P’s subsequently heating up. The first
relationship is given by a certain sequence of propositions, and the second is given by what is
merely one small segment of that sequence.
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

Before we discuss why the Jackson-Petit attempt to save functionalism doesn’t go

through, let us briefly review why functionalism is prima facie inconsistent with the
fact that mental states have causal properties. At time t, I have an intense desire to
drink ice-water. Let B be the brain-state that realizes that desire. At that same time, I
see a glass of ice-water in front of me. Because I am thirsty, I grab it and drink its con-
tents. It is intuitively clear that, at t, it is my desire for ice-water that causes me to drink
the contents of the glass. So my desire for ice-water is active at t. Thus B’s being a desire
for ice-water is active at that time. If functionalism is right, then B’s being such a desire
supervenes on its functional role. (Let R be that role.) But B’s having such a role is not
causally potent at t. The expression “B’s having functional role R” denotes an abstract
property of a sequence of events. In this respect, it is less like the expression “the
squareness of the peg” than it is like the expression “the fact that people gravitate to-
wards extreme political positions in crisis-conditions.” Functionalism cannot accom-
modate the fact that, at t, my desire for ice-water is doing any work.
Jackson and Pettit deal with this by saying the following. First of all, given that B
has R, it is guaranteed that, at t, I will grab the glass and drink its contents. B’s having
R therefore programs for my grabbing the glass and then drinking its contents. Thus,
Jackson and Pettit conclude, functionalism allows mental states to be program-causes,
and therefore is compatible with the causal potency of the mental.
But this argument overlooks one crucial fact. The just-mentioned guarantee is
logical, not causal. Given that B has R, it follows tautologously (and therefore analyti-
cally) that I will drink from the glass. But causal relations are not expressed by tautolo-
gies. Given the statement there is a flame underneath the pan, the statement the pan will
heat up doesn’t follow tautologously (or otherwise analytically) and that is why the first
proposition can provide a causal explanation for the truth of the second. By the same
token, B has R provides no causal explanation for the truth of JM drinks from the glass,
given that the latter is an analytic consequence of the former.

The first relationship has a very complicated structure. It starts with a generalization. It then
transitions to two singular propositions. From each of these singular propositions, it derives
what must be regarded as a program-explanation, given Jackson and Pettit’s own explicit state-
ments. (So we are dealing with program-explanations embedded within what is supposedly a
program-explanation.) From each of these two embedded program-explanations, it derives two
cases of explanation in terms of effective-causes. Finally, on the basis of these last two explana-
tions, it logically derives another generalization.
The embedded program-explanations obviously don’t have the same structure as the expla-
nation of (PA) in terms of (PH). In fact, it seems that, on pain of vicious regressiveness, they
couldn’t have that structure. (But, as I said before, I will not discuss this last point.) What is clear
is that Jackson and Pettit have over-extended the term “program-cause” in using it to apply to
both kinds of relationships.
 Conceptual Atomism and the Computational Theory of Mind

Summary of our findings thus far

The concepts of a program-cause and of program explanation are of the highest impor-
tance. But for program-explanations, explanation outside the arena of microphysics would
be impossible (Also, for reasons that I haven’t given, but that Jackson and Pettit (2004d) do
give, explanations in terms of program-causes are in some ways more informative than
explanations in terms of effective causes. Given a biological description of the events that
led to Smith’s death, i.e. given a description involving concepts such as arrhythmia and left
ventricular hypertrophy, it can be inferred how Smith’s death could have been avoided (he
shouldn’t have played soccer after taking amphetamines). But that very information cannot
be inferred from a description, even a maximally precise one, of the subatomic events that
led to Smith’s death. The biological explanation thus bears counterfactual information
about the events in question not borne by the microphysical explanation. Counterfactual
claims about a situation are warranted only where there is a knowledge of the laws govern-
ing that situation. (Unless you have some understanding, however obscure and primitive,
of the forces involved in the rock’s shattering the window, you have no justification for af-
firming the counterfactual that the window wouldn’t be broken if the rock hadn’t collided
with it.) Consequently, the biological description bears information, not borne by the cor-
responding microphysical description, as to the nature of the laws governing those events
and, therefore, as to the principles in terms of which those events are to be understood. For
these reasons, a biological description of a given event is (in some contexts, in some re-
spects) more informative than the corresponding microphysical description. In general,
higher-level descriptions of events provide some information not carried by the corre-
sponding lower-level descriptions.)
A paradigmatic program-cause is the case of the pan’s heating in consequence of
the flame’s being put underneath it. The concept illustrated by this paradigm is an im-
portant one. This concept is instantiated by ordered pairs of states of affairs such that
the first state of affairs is literally composed of the events that effectively cause the sec-
ond state of affairs.
But we render the term “program cause” ambiguous, or lacking in any coherent
meaning, if we use that term to denote relations that don’t hold between states of af-
fairs. Noun-phrases – even non-empty ones – don’t always denote states of affairs. The
expression:
(%) “the fact that, throughout time, different cultures have always had different
ways of indicating differences in social status”
is not empty. But (%) isn’t comparable to “the flame underneath the pan”, “the fact that
the water is 112°”, or “the squareness of the peg.” In their attempts to reconcile content-
externalism and functionalism with the causal potency of the mental, Jackson and
Pettit use the term “program-explanation” to denote relations that don’t hold between
states of affairs – they use it to refer to the relations that hold between “entities” like
that denoted by (%). For that reason, those attempts are fallacious.
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

Of course, given only that Jackson and Pettit didn’t succeed in their efforts, it
doesn’t follow that functionalism and content-externalism are in fact incapable of ac-
commodating the causal potency of the mental. But we have seen that those doctrines
do indeed have exactly that defect. For reasons already given, it follows that content-
externalism is false; and this fact confirms the positive doctrines that we have put forth
concerning mental content.

Some additional objections to functionalism

Because functionalism is practically an orthodoxy in contemporary philosophy of

mind, it is worth our while to state some additional arguments against it, so as to rein-
force our conclusion that it is false. We will see that the objections to functionalism
raised by these additional arguments are naturally understood as corollaries of the fact
that functionalism is incompatible with the causal potency of the mental.
Suppose that, because he has brain-state B, Smith says “three” when asked “what
is 1+2?” Even though B is what mediates between perceptual input (his hearing the
question “what is 1+2?”) and behavioral output (his saying “three”), it doesn’t follow
that any arithmetical belief of his was involved. It could be that, because of B’s electrical
properties, the sounds he heard triggered an involuntary laryngeal spasm which led to
his producing the sound “three.”
An analogous point holds with regard to any other example of a case where B
mediates between cause and effect in a way that is characteristic of a belief that 1+2=3.
One day Smith learns that Jones has one car and two motorcycles, and no other motor-
ized vehicles. For some reason, he finds this knowledge very exciting. This causes his
heart to race. Given B’s electrical (not its representational) properties, and given some
quirk in Smith’s neural architecture, the racing of his heart causes Smith to form the
belief Jones has three motorized vehicles. Given the input (Smith’s learning that Jones
has two motorcycles and a car), B leads to the right output (Smith’s believing that Jones
has three motorized vehicle). In other words, given the input, B leads to the same out-
put as a belief that 1+2=3.
But we clearly don’t have a case where Smith’s believing that 1+2=3 led to his hav-
ing the right belief. We have a situation where dumb luck simulated the operation of
such a belief. In general, if B mediates between input and output in the wrong way, then
B isn’t a belief that 1+2=3, regardless of whether B consistently assigns the right output
to any given input. And for B to mediate between input and output in the right way is
for B’s being a certain kind of belief to be what generates input-appropriate outputs.
But, of course, if B’s being a certain kind of belief is what is responsible for B’s generat-
ing the right outputs, then functional role is a causal effect of representational content.
And, in that case, B’s having that representational content isn’t constituted by its having
that functional role.
 Conceptual Atomism and the Computational Theory of Mind

In general, so far as content is responsible for causal role, content cannot be identi-
fied with causal role. But where content isn’t responsible for causal role, there isn’t a
case of anything’s being a belief (or an intention or a regret...) doing anything: in other
words, we don’t have a case where anything mental is doing anything.
The functionalist has a response to this:
I grant that freak accidents, such as the ones you’re describing, might happen. But
if B always leads to the right output, the chances that B is anything other than a
belief that 1+2=3 are vanishingly close to zero.

That is entirely correct. But we mustn’t confuse issues relating to what B is with issues
relating to how we ascertain what B is. Supposing that B always produces the right
output, that is excellent evidence that B is a belief that 1+2=3. But we’ve just seen that,
given any instance where B generates the right output, B’s generating that output can,
in principle, be explained without supposing that B realizes a belief that 1+2=3. Given
any number of instances where B generates the right output, that fact can in principle
be explained without having to suppose that B realizes a belief that 1+2=3. This means
that the statement B mediates between input and output in a way characteristic of a be-
lief that 1+2=3 does not analytically entail, and is not analytically entailed by, B is a
belief that 1+2=3.
Of course, the objector is right that, if B mediates between input and output in a
way that is characteristic of such a belief, then refusing to characterize B as such a be-
lief may be nothing more than childish skepticism. But given that such skepticism is
not incoherent and that it is not analytically false, it follows that B’s mediating between
input and output in the right way is merely evidence of B’s being a belief that 1+2=3,
and is thus not constitutive of its being such a belief.
There is another, similar argument against functionalism.7 Suppose that B assigns
the right output to a thousand different inputs. (For example, given that Smith has just
heard “Brown has 1+2 bicycles”, B leads Smith to say “therefore Brown has 3 bicycles.”)
Given the evidence, it is likely that B is a belief that 1+2=3. But that same evidence is
consistent with the hypothesis that B is a belief in some “bent” proposition, e.g. 1+2=3
– except when the objects being added are insects: in that case, 1+2=4. So as long Smith
hears or reads anything not having to do with insects, B will assign the same output to
any given input as a belief that 1+2=3. But if Smith hears, for example, “there are two
cockroaches in Brown’s room, and one ant, and no other insects”, Smith says “therefore
there are four insects in Brown’s room.”8
In general, no matter how often B generates the right output, it is a possibility that
B mediates a belief in some proposition other than 1+2=3. Given that this is even a

7. This argument is similar to one made by Kripke (1982) and also to one made by Wittgens-
tein (1958). I will set aside the question whether it actually coincides with anything either of
these authors said. See also Lepore (1994: 197-199).
8. See Kripke (1982), Blackburn (1984: Chapter 2).
 Chapter 12.  Jackson and Pettit on program-causality and content-externalism 

skeptical possibility, it follows that B’s having the same functional role as a belief that
1+2=3 (i.e. its assigning the same output to any given input as such a belief) is not
constitutive of B’s being such a belief and is at best good evidence to that effect. This
confirms our earlier point that, if B really is a belief that 1+2=3, then B’s assigning the
right output to any given input is at most an effect of its being such a belief. After all, X
is evidence of Y only if X is an effect of Y (or, in any case, only if X stands in some kind
of causal relation to Y).
Let us sum up. Even if B produces the right output for any input, it doesn’t follow
that B’s being a belief that 1+2=3 is involved. But so far as B’s being such a belief is in-
volved, functional role is a veritable effect of content, and is thus not constitutive of it.
Thus, either (i) B’s functional role is a causal consequence of its content, and is therefore
distinct from it, or (ii) B’s functional role is consistent with B’s being any number of be-
liefs distinct from a belief that 1+2=3. Both options are unacceptable.9 Given (i), it im-
mediately follows that functionalism is false. Given (ii), it follows that functionalism is
incapable of recognizing the distinction between believing that 1+2=3, on the one hand,
and believing some bent variant of that proposition, on the other. Since there obviously
is such a difference, we have yet another reason to conclude that functionalism is false.

9. Conceivably, one could say that, because one’s dispositions are consistent with one’s having
any number of different beliefs, we should hold that it is “indeterminate” what our beliefs really are.
So supposing that + A, + B, and + C are “bent” versions of addition, Smith believes 1+A2=3, 1+B2=3,
1+C2=3, and so on, no less than he believes 1+2=3. Given such a view, it is a short step to advoca-
ting a kind of anti-realism about propositional attitudes. Since I obviously don’t believe 1+B2=3,
then supposing that I believe that proposition no less than I believe 1+2=3, it follows that I don’t
believe the latter. Given any apparent instance of belief, it is obvious how to generalize this line of
thought so as to demonstrate the non-existence of that belief. In general, one could take functio-
nalism as demonstrating the need for an anti-realist view of propositional attitudes. Wittgenstein
(1958) and Quine (1960) seem to have gone exactly this route.
But it is incoherent to deny that there are propositional attitudes, since such a denial itself
embodies a propositional attitude. (This argument is not a new one. Versions of it are found in
Strawson 1994 and Hacker 1996. Churchland 1981 argues that it is question-begging.) In any
case, it is suspicious that functionalism has such revisionist consequences; and I am taking it for
granted that people do have beliefs, aspirations, and so on.
 Part II

Fodor, conceptual atomism, and

computationalism

In the preceding pages, we put forth some very definite views as to the nature of con-
ception. Those views are incompatible with a powerfully argued analysis put forth by
Jerry Fodor. In this Part, we will evaluate Fodor’s analysis.
Fodor’s work makes it clear that content-externalism has a great deal of independ-
ent support. So in order to prove the mettle of our own analysis, it is necessary that we
consider Fodor’s alternative to it in detail. Doing so will give us some insight into a
number of important questions: Are we computers? Supposing that we are, what does
that entail? Does it entail, for example, that thinking consists in operations on linguis-
tic symbols? Supposing that it does, what is the nature of those operations? Are those
operations to be understood exclusively in terms of the purely formal properties of the
symbols involved, or are they better understood in terms of their semantic properties?
Or can they be fruitfully understood in both ways? (In other words, does the last ques-
tion embody a false dichotomy of some kind?) These are but a few of the questions that
we will address.
 chapter 13

Content-externalism and conceptual atomism

We must begin by discussing a doctrine that we will refer to as “conceptual atomism.”

Some background is needed before we can say what that doctrine is.
According to the view that we have defended, for any spatiotemporal entity x, my
having a concept of x involves my knowing some description that x satisfies. An im-
mediate consequence of this view would seem to be that in order to have a concept of
any spatiotemporal object, one must already have concepts. Where such objects are
concerned, there is no possibility of conception in a vacuum of other concepts.
Of course, not everybody will agree that our particular view of conception is the
right one. But what is generally, though not universally, accepted is that possession of any
one concept presupposes possession of others.1 As we discussed earlier, it seems prepos-
terous to assume that an otherwise conceptless creature could have a concept of Socrates
or of the Washington Monument. It seems even more preposterous to assume that an
otherwise conceptless creature could have thoughts about theoretical entities, such as
electrons or neutrinos. There doesn’t seem to be any spatiotemporal entity x such that
one could have a concept of x without having a concept of anything other than x. The
idea that one’s x-concept could exist in isolation thus seems to be a non-starter (Berkeley
1934, Sellars 1963, Bonjour 1985, Peacocke 1992, Kuczynski 2003, 2004).
Concepts are not atoms; they are not self-contained units. Where there is one con-
cept, there are many. Conceptual atomism is false, and conceptual non-atomism – con-
ceptual molecularism or holism – must therefore be correct. In any case, this seems like
a reasonable position; and our analysis is consistent with it, since it makes it a precondi-
tion for my having any concept of anything that I have a number of other concepts.
Here we must make a distinction. In this work, we have discussed what it is to have
a concept of a spatiotemporal entity. We have not discussed what it is to have a concept
of a non-spatiotemporal entity – of a property or a proposition. So we have left it open
whether conceptual atomism holds with respect to our concepts of abstracta. We will
discuss this problem in Chapters 22–23. We will find that non-atomism is no more true
with respect to our concepts of abstracta than it is with respect to our concepts of spa-
tiotemporal entities. (Actually, in those chapters, we will find that the distinction be-
tween those two kinds of concepts is much blurrier than has thus far been generally as-

1. See Bonjour (1985) and Stich (1983). Sellars (1963) and Wittgenstein (1958, 1975) deserve
credit for giving modern epistemology its generally non-atomist orientation. But, as I have pre-
viously said, it is really in George Berkeley’s book A Theory of Vision that one finds the earliest,
and also the most cogent and perspicuous, defense of non-atomism.
 Conceptual Atomism and the Computational Theory of Mind

sumed. For we will find that many so-called “constituents” of the spatiotemporal world
– e.g. rocks, trees, vases – are more abstract than we are generally inclined to think.)
But independently of any theories that we will put forth, there are some obvious
and compelling reasons to reject an atomist conception of conception. Could one grasp
the concept of justice without grasping concepts like sentience, awareness, and morality?
And could one grasp any one of these concepts without grasping others? Surely one
couldn’t have any conception of morality unless one understood what it was for an as-
piration to be thwarted or, more generally, for a sentient being to be violated. So a con-
cept of morality would seem to presuppose concepts of purposiveness, frustration, and
sentience. And to have a grasp of any one of these concepts, one must have yet others.
To grasp the concept of frustration, one must grasp the concept of desire; and to grasp
the concept of desire, one must grasp the concept of an attitude in general.
It used to be thought that, sooner or later, this regress would terminate in founda-
tional concepts (Russell 19842, Wittgenstein 1922). But the attempts to find such foun-
dational concepts have borne no fruit, and philosophy has generated a number of
positive reasons for thinking them not to exist (Sellars 1963, Wittgenstein 1974, 1983,
Kuczynski 2006). We will discuss these in due course.
In any case, given only the remarks just made, it is hard to believe that a creature
could grasp one abstractum without grasping a multiplicity of others. Conceptual at-
omism would thus seem to be as false with respect to our grasp of such entities as it is
with respect to our grasp of spatiotemporal entities.

Some reasons to accept atomism

Fodor has made it clear that, if one accepts content-externalism, one must advocate an
atomistic conception of what it is to be aware of spatiotemporal objects. Fodor accepts
content-externalism, and he thus accepts conceptual atomism. Even though many
agree with Fodor that content-externalism is correct, few agree with Fodor’s concep-
tual atomism. But, so far as one is a content-externalist, one is guilty of incoherence in
rejecting Fodor’s atomism.
As we will see, Fodor is an atomist across the board: conceptual atomism holds
with regard to our grasp of both spatiotemporal and abstract entities. In Fodor’s view,
an otherwise conceptless creature could grasp the concept of justice or number or
money (1990, 1998). A creature that didn’t have concepts like society or property could,
according to Fodor, have the concept of money.
Of course, this is, on the face of it, an implausible view. But to echo what we said a
moment ago, content-externalism demands that just this view be accepted. So supposing
that content-externalism is correct, Fodor is guilty only of seeing what others fail to see.

2. This work was written in 1914. Under pressure from Wittgenstein, Russell decided not to
publish it, and it thus wasn’t published until 1984.
 Chapter 13.  Content-externalism and atomism 

Fodor realizes that conceptual atomism is a counterintuitive doctrine. For this

reason, he provides a number of brilliant arguments on behalf of it (Fodor 1998). He
also successfully shows that a number of independently plausible positions actually
demand an acceptance of atomism. (He does not, however, show that those indepen-
dently plausible positions are correct. And we will see that they are not correct.)

Why content-externalism presupposes conceptual atomism

In this section, I would like to discuss why content-externalism demands an atomistic

conception of conception. In this context, we will suppose for the sake of argument
that content-externalism is correct.
According to that doctrine, the content of a mental state is a function at least in
part of the origins of that state. Brainstates b and b* may have different contents even
if, leaving aside their causal origins, they are qualitatively identical. For example, b
may be a concept of Smith, while b* is a concept of Jones, even though (leaving aside
their distal causes) b and b* are qualitatively identical, and are embedded in the same
way in qualitatively identical nervous systems.
It may help to recall the story that we told in Chapter 1. The information encoded in
Max’s perceptions and subsequent thoughts is different from that encoded in Twin-Max’s
corresponding perceptions and thoughts; and those differences lie entirely in the fact that
the one person’s mental states have different origins from the others. They lie entirely in
the fact that, beginning at time t, Max’s perceptions and subsequent thoughts do, where-
as Twin-Max’s do not, originate with some state of affairs that have rock R as a constitu-
ent. It isn’t as though, in order for Max and Twin-Max to diverge in this way after time t,
they must already differ in respect of what concepts they have. The just-described differ-
ences between Max and Twin-Max don’t presuppose any conceptual differences.
In general, if content-externalism is right, two creatures can differ with respect to
a single concept. X and Y can be absolutely identical except that, where X has a concept
of Smith, Y has a concept of Jones. So it is a straightforward consequence of content-
externalism that some concepts are atoms.
A corollary is that atomism is false only to the extent that the content of a mental
state is not a function of its causal origins. On pain of incoherence, one is an atomist to
the extent that one is a content-externalist. Content-externalism is a form of atomism.
Conceivably, one could be an externalist with regard to certain concepts (e.g. one’s
concept of Smith) but a non-externalist with regard to other concepts (e.g. one’s con-
cept of the number five). Indeed, I am sure that this is the position that most self-de-
scribed content-externalists would take. If Smith and Jones are atom-for-atom dupli-
cates of each other, then surely Smith has a concept of the number five iff the same is
true of Jones. Content-externalism is presumably meant to hold only with respect to
our concepts of spatiotemporal objects.
 Conceptual Atomism and the Computational Theory of Mind

But even if we confine ourselves to this class of objects, it is pretty clear that con-
tent-externalism is not meant to hold across the board. Any theoretical notion is ipso
facto one that can be grasped only in the context of a number of other notions (Hem-
pel 1952, 1965, Nagel 1962, Carnap 1966). One couldn’t have a concept of a quark un-
less one had some understanding of the structure of the atom and, therewith, of the
strong and weak forces, and so on. This is not to mention that one would have to have
at least some grasp of concepts such as motion, energy, mass, time and the like. A being
that had no concept of motion surely wouldn’t have a concept of a quark. Thus, one
grasps the concept of a quark in terms of these other notions. This suggests that one has
a concept of a quark only if one knows the truth of some statement that describes quarks
in terms of these other concepts. If this is right, as I believe it is, then having a concept
of a quark involves knowing a true proposition that describes quarks. Since any such
proposition would inevitably be composed of a multiplicity of concepts, it follows that
one couldn’t have a concept of a quark without having many other concepts.
What we said about a concept of a quark is true of a concept of anything spatio-
temporal, whether theoretical or not. To the extent that one’s concept of Smith is de-
scriptive, it cannot be understood in atomist, or therefore in strictly externalist, terms.
So the content-externalist must hold that one’s concept of Smith is an atom.
A moment ago we said that one’s knowledge of theoretical entities is essentially
descriptive. This is not to advocate some kind of instrumentalism or anti-realism about
theoretical entities. A good case has been made that instrumentalism, operationalism,
and other forms of theoretical anti-realism are quite erroneous (Hempel 1952: 29–50,
1965: 123–135, Pap 1963: 39–743). But given only that theoretical entities are to be
understood in terms of non-theoretical data, it doesn’t follow that such entities are
nothing but constructs of such data. To think otherwise is to fail to make some basic
distinctions relating to scope.
Remember what we said about the word “Socrates.” When one is told:
(i) “Socrates” is the name of the great philosopher who drank hemlock,
what one is being told is not:
(ii) “Socrates” names anything x such that x is a unique great philosopher who
drank hemlock (and exactly one thing was a great philosopher who drank
hemlock).
Rather, what one is being told is:
(iii) Somebody x is a unique great philosopher who drank hemlock, and “Socra-
tes” names x.

3. Bas van Fraassen (1982) revived a version of anti-realism with regard to theoretical entities.
 Chapter 13.  Content-externalism and atomism 

The descriptive information is given wide-scope with respect to the “names”-operator, so

that it has a reference-fixing, as opposed to a meaning-giving, role. But the term “Socrates”
is still understood in terms of the description to the left of the “names”-operator.
Similarly, when one is told:
(i*) the term “quark” refers to instances of that natural kind whose supposed exist-
ence helps account for data D1…Dn,
one is not being told:
(ii*) the term “quark” refers to instances of any unique natural kind N such that the
supposed existence of instances of N helps account for D1…Dn.
Rather, one is being told:
(iii*) there is some unique natural kind N such that the supposed existence of in-
stances of N helps account for D1…Dn and such that the term “quark” refers
to such instances.
So some description referring to D1…Dn fixes the referent, but doesn’t give the mean-
ing, of the term “quark.”
Given this fact, it is obvious why there is no tension between, on the one hand, the
obvious truth that theoretical entities are understood in terms of non-theoretical data
and, on the other hand, the well-established thesis that statements about theoretical
entities are not synonymous, or otherwise analytically equivalent with, statements
about macroscopic data.4 Because that data has a reference-fixing role, there are coun-
terfactual scenarios where those same theoretical entities generate very different ex-
perimental data. So the counterfactual properties of statements about that experimen-
tal data don’t track the counterfactual properties of statements about those theoretical
entities; and this thwarts attempts to reduce the one set of statements to the other.

Fodor’s reasons for accepting conceptual atomism

Fodor sees very clearly that conceptual atomism is a direct consequence of content-
externalism. So given how implausible atomism is, why doesn’t Fodor reject content-
externalism?
Before content-externalism even existed5, Fodor (1968, 1975) advocated the so-
called Computational Theory of Mind (CTM). As Fodor has always realized, CTM de-

4. See Hempel (1952: 23-39) for what are generally considered to be cogent refutations of the
thesis that hypotheses about imperceptible entities are reducible to statements about the macros-
copic or the phenomenal.
5. Or, at any rate, before it was a widely accepted doctrine.
 Conceptual Atomism and the Computational Theory of Mind

mands an acceptance of conceptual atomism. So Fodor correctly sees CTM as providing

independent corroboration for the atomism demanded by content-externalism.
Fodor also advocates the Symbolic Theory of Thought (SCT). As Fodor realizes,
CTM demands SCT. But Fodor also sees clearly that, apart from its connection with
CTM, there is much support for SCT. So Fodor correctly sees SCT as providing inde-
pendent corroboration for the atomism demanded by content-externalism.
As previously noted, Fodor’s arguments for CTM are intertwined with his argu-
ments for SCT. Let us now consider those arguments

The raison d’être for CTM

CTM is given by the thesis:

(*) “To think is to compute; thinking is computing.”6
Before we consider the reasons for accepting CTM, we must ask: what does the word
“compute” mean? What is it to perform a computation? Right now I could perform a
computation; I could compute a sum (i.e. I could add two numbers). But given that
interpretation of the word “compute”, (*) is patently false, at least as a general statement
about cognitive activity. This is because a precondition for my adding two numbers is
that I already have many thoughts, and that I thus already have a wealth of concepts. I
must know various symbolic conventions (e.g. that “1” denotes the number one); I
must grasp various mathematical principles; I must have cognitive wherewithal to
have perceptions of enduring physical objects (e.g. perceptions of the paper and pencil
that I will use to compute the sum in question) and also to synthesize those percep-
tions into a body of mnemic information on which I can draw as circumstances re-
quire (I won’t be able to add if I can’t keep track of the various figures I write down, or
of what my reason for doing so was).
So if the word “compute” is taken in its customary sense – in the sense that it bears
in statements like “Smith is performing a computation” – then the existence of compu-
tations is actually an extremely derivative form of mental activity and, consequently,
CTM is viciously circular.
Fodor is aware of this problem, as are most advocates of CTM. In an attempt to
avoid this vicious circularity, advocates of CTM define the terms “compute” and “com-
putation” as follows:
(FC) Every case of computing is a case of manipulating symbols, but not vice versa.
For a symbol-manipulation to qualify as a computation, it cannot be to any

6. Fodor (1987: 18-20).

 Chapter 13.  Content-externalism and atomism 

degree semantics-driven, and must therefore be driven entirely by the forms of

the symbols in question.7
When a calculator adds a sum, it doesn’t consider what is meant or referred to by the
discolorations on its monitor; the calculator’s behavior is driven entirely by the formal
properties of those symbols. A computation is thus a form-driven manipulation of
symbols. When the words “compute” and “computation” are taken in this sense, there
is no vicious circularity in saying that the mind is a computer, and that cognitive activ-
ity consists entirely in the performing of computations. So CTM is given by (*), with
the qualification that the words “computation” and “compute” are to be understood in
the sense assigned to them by (FC).

Some commitments of CTM

As we will see in a moment, there are excellent reasons to advocate CTM. But before
we consider those reasons, we must make a couple of points explicit. CTM is obvi-
ously committed to the idea that we think in symbols – that the basic units of thought
are expressions and that, consequently, we think in a language.
A consequence is that CTM is committed to conceptual atomism. If, as CTM re-
quires, we think in a language, then its expressions are our concepts. Any language
must have ultimate units of significance (also known as “morphemes”). Given any lan-
guage, some of its expressions are complex, and others (“Socrates”) are simple. It isn’t
possible for every expression of a language to be complex. So supposing that we think
in a language, some of the expressions of that language are simple and foundational,
and all the rest are built out of those simples. Of course, if we think in a language, then
our concepts are symbols of that language. It follows that, if we think in a language, then
some of our concepts are simple and foundational – they are atoms – and the rest are
built out of those atoms. CTM is thus committed to conceptual atomism.
There is a possible objection to this:
You say that, if we think in a language, then some of our concepts are founda-
tional and therefore simple. But this follows only if one rejects a certain plausible
view as to the nature of linguistic meaning. According to that view, the meaning
of an expression is constituted by its inferential role. Given any sentence of the
form “Aristotle was wise” that sentence follows from certain other sentences (e.g.
“either there are square circles or Aristotle was wise”) and certain other sentences
follow from it (e.g. “somebody was wise”). In general, given any sentence of the
form F…Aristotle…G, certain sentences entail that sentence, and that sentence en-
tails certain other sentences; and what “Aristotle” means is a function entirely of
what those entailment relations are. Even more generally, for any expression E,

7. Fodor (1981: 13-17, 226-227, 1987: 18-20, 1994: 294-298).

 Conceptual Atomism and the Computational Theory of Mind

any sentence of the from F…E…G entails, and is entailed by, certain other sen-
tences; and what E means is a function of what those various entailment-relations
are. If you accept a conceptual role account of meaning – if you accept “Concep-
tual Role Semantics” (CRS) – then the meaning of any given symbol is determined
by its relations to other symbols; and, from a semantic viewpoint, there are thus
no “simple” or “foundational” symbols. As we just saw, supposing that “Aristotle”
has meaning (or reference) M, it has that meaning in virtue of the fact that “Aris-
totle” is related in a certain way to other symbols. So there is no straightforward
inference from “we think in language” to “some kind of conceptual atomism
holds” (Sellars 1963, Field 1977, Brandom 1998).

There are several points to make here. The first of these points is made by Fodor himself
(Fodor 1998: 138, Fodor and Lepore 202: 1019). If one accepts CTM, one cannot advo-
cate any form of CRS. CTM is trying to account for cognitive behavior in terms of op-
erations on symbols. According to CRS, something can be a symbol only if it is already
a part of a pattern of inference-making. So CRS makes the having of vast networks of
thoughts be a prerequisite to the use of any symbols. Thus, to avoid vicious circularity,
any identification of thought with the manipulation of symbols must not accept a CRS-
account of meaning; and any such identification must accept the conventional view that
some of the symbols of any given language are foundational and atomic, and are thus
not to be defined in terms of other symbols belonging to that language.
Consider the sentences “Socrates snored” and “a great philosopher who never
wrote down any of his ideas snored.” We know from Kripke that the second sentence
does not follow analytically from the first. If, in the first sentence, we replace “Socrates”
with “Aristotle”, the sentence that results won’t entail the other. In general, uniformly
replacing occurrences of “Socrates” with “Aristotle” (or vice versa) won’t turn a non-
entailment into an entailment. Now consider the sentences “Socrates snores” and
“something snores.” The first sentence does entail the second. If, in the first sentence,
we replace “Socrates” with “Aristotle”, the sentence that results will also entail “some-
thing snores.” In general, uniformly replacing occurrences of “Socrates” with occur-
rences of “Aristotle” (and vice versa) will never turn an entailment into a non-entail-
ment. This shows that, given only what the entailment-relations are of sentences
containing “Socrates” and of sentences containing “Aristotle”, there is no reason to sup-
pose that those names do not co-refer. So CRS cannot accommodate the fact that “Ar-

8. Fodor (1998: 13) writes: “For fear of circularity, I can’t both tell a computational story about
inferences and tell an inferential story about what content is. Prima facie, at least, if I buy into
Inferential Role Semantics, I undermine my theory of thinking.”
9. Fodor and Lepore (2002) write: “Inferentialism is prohibited from taking this route, since
the inferentialist’s point in a nutshell is that meaning is a construct from inferential role, not the
other way around.”
 Chapter 13.  Content-externalism and atomism 

istotle” and “Socrates” refer to different things and are therefore semantically different.
Therefore CRS is false.10
One could attempt to side-step this last point by saying that, given what we’ve
learned about Aristotle and Socrates, “Aristotle snored” does have different entailment-
relations from “Socrates snored.” Given our historical knowledge, the one does, while
the other does not, entail “the author of the Metaphysics snored”; and the one does,
while the other does not, entail “somebody who died of hemlock-poisoning snored.”
But, as Fodor and Lepore (2002: 101) say, if we take that point of view, then every true
sentence of the form FAristotle has phiG becomes an analytic truth. The sentence “Aristotle
died before his 90th birthday” becomes analytic. Since it obviously isn’t analytic, the point
of view just described is false. Since CRS is false, we may conclude that SCT is committed
to conceptual atomism.
Conceivably, an advocate of SCT could counter-respond by saying that the Men-
talese symbol for Aristotle is different from its counterpart in English: the former is,
whereas the latter is not, to be understood in terms of its inferential role. But given this
position, one’s Mentalese symbol for Aristotle becomes so different from “Aristotle”
that it can no longer be regarded as a mere translation of the latter. More generally, this
position makes the so-called “symbols” of Mentalese so different from their English
counterparts that it becomes unclear whether the former are fruitfully likened to lin-
guistic expressions. This is an issue that will recur throughout the remainder of this
work. In any case, we may conclude that Fodor is entirely right to hold that if SCT is
correct, then so is conceptual atomism.

How CTM solves the tracking-problem

Let us now turn to the question: Why advocate CTM? (The argument about to be
given is found in Fodor 1981: 13–17, 226–227, 1987: 18–20, 1994: 294–298.) Thoughts
are identical with, or at least realized by, brain-activity. (In any case, this is a plausible
and widely held view. Given that the mental and the physical interact, and given that
physical realm is explanatorily closed, i.e. that physical events never have to be ex-
plained in terms of non-physical events, it is de rigueur to suppose that the mental is
realized by the physical.) Brains are physical objects, and are therefore governed by the
laws of physics. The laws of physics are not laws of rationality. Whether the billiard ball
goes in this as opposed to that direction has nothing to do with canons of good reason-
ing. The same is true of our brains and of the various structures they comprise. All of
these things are entirely physics-driven, and there would thus seem to be no room for
canons of good reasoning to play any part.
At the same time, our cognitive activity is rational, at least to a high-degree. Our
thoughts follow one another in ways that are generally consistent with the canons of

10. Kripke (1972), Fodor and Lepore (2002).

 Conceptual Atomism and the Computational Theory of Mind

logical reasoning. But how is this possible, given that those thought-sequences are
purely physics-driven and that the laws of physics have nothing to do with the canons
of good reasoning? How is it possible for rationality to track physics, given that every-
thing spatiotemporal falls within the scope of physical law and that physical laws are
not laws of rationality? We will refer to this as the “tracking-problem.”
CTM solves this problem. Symbols (or symbol-tokens, strictly speaking) are pure-
ly physical objects. But we can create environments in which they follow one another
in ways that are consistent with the strictures of logic. Given a set of symbol-tokens, we
can create an environment such that, given the laws of physics and given the physical
properties of those symbol-tokens, the latter are physically compelled to generate oth-
er symbols such that the resulting sequence of symbols constitutes an instance of ratio-
nal ideation (Fodor 1981: 226, 1981b, 1987: 18–20, 1990: 22).
Consider the symbols “Jerry is not bald”, “either Smith is tall or Jerry is bald”, and
“Smith is tall.” We can create a physical environment such that, if the first two symbols
are tokened (within a given period of time) in that environment, then the third symbol
is tokened in that same environment shortly thereafter.
Here is another example. Consider the symbols “12”, “+”, “14”, “=”, “26”. We can create
a physical environment such that, if the first four symbols are tokened (in that order) in
that environment, then the fifth is tokened in that same environment shortly thereafter.
In fact, not only can such environments be created: we create them all the time.
They are called “computers” or “calculators.” (In this context, the distinction between
calculators and computers isn’t important.) So the tracking-problem vanishes if we
assume that brains are computers.

CTM incompatible with the causal potency of representational content

There is an obvious problem with CTM – but if one accepts content-externalism, that
apparent problem turns out to be a virtue. Contrary to what CTM says, our thinking
obviously is (or at least seems to be) semantics-driven. I believe that Smith is either in
Iceland or in Jamaica. I then learn that he is not in Jamaica. Soon thereafter, I form the
belief that he is in Iceland. Surely it is (or at least could be) because the first two thoughts
have certain contents that I end up having the third thought. So supposing, as CTM does,
that my having these thoughts consists in various expressions being tokened in my brain,
the sequences of expressions in which my thinking consists are content-driven, and not
(at least not exclusively) form-driven. Thus, if indeed we think in symbols, our thinking
is semantics-driven, not form-driven, and CTM is therefore false.
Here there are two points to make. First, as we’ve already seen, if content-external-
ism is right, then content is causally inert. What is causally potent is always what is
local; and, according to content-externalism, a brain-state’s having a certain content
does not supervene on what is local: on the contrary, it supervenes specifically on what
is non-local – what is distal and purely relational and therefore causally hollow. So, as
 Chapter 13.  Content-externalism and atomism 

Fodor (1981, 1987b) clearly says, content-externalism is incompatible with the view
that semantics is causally operative. Thus, given an acceptance of content-externalism,
it is de rigueur to hold that thought-processes are not content-driven. Thus, if content-
externalism is correct, then it is to the credit – not the discredit – of CTM that strips
content of causal potency. We may conclude that, given his acceptance of content-ex-
ternalism, Fodor is entirely right not to hold it against CTM that it is inconsistent with
the presumed causal potency of mental content.
Further, for a theory to be adequate, it must be consistent with the relevant data,
and not with our a priori beliefs (except in so far as our a priori beliefs are consistent
with the data). In this context, the data is that certain kinds of thoughts are likely to
follow certain other kinds of thoughts. Just as there are regularities in the non-mental
world, so there are regularities in the mental world; and the latter regularities are the
data that a correct model of cognition must accommodate. It is not a datum that se-
mantic content is what is responsible for those regularities. As Hume pointed out, it is
not a datum that there are “forces” or “powers” in the physical world. Similarly, it is not
a datum that, in the cognitive sphere, there are “semantic forces.” This isn’t to say that
such forces don’t exist, but only that the question whether such forces exist must be
answered on the basis of empirical research. If Fodor is right, such research has estab-
lished that they don’t exist.
CTM thus solves the tracking-problem. And supposing (as Fodor and many oth-
ers do) that content-externalism is correct, and therefore that the mental is causally
impotent, the fact that CTM strips semantic-content of causal powers actually re-
dounds to its credit.
This is not to mention the fact that CTM provides a speedy solution to the mind-
body problem. There is obviously no difficulty explaining how computers arise out of
non-mental, purely physical entities. So supposing that there our brains are comput-
ers, there is no difficulty explaining how mind arises out of such entities. There is
therefore much to be said for CTM.
We’ve seen that an acceptance of CTM demands an acceptance of SCT; and we’ve
seen that, at least in this context, an acceptance of SCT demands an acceptance of
conceptual atomism. So there is much to be said for conceptual atomism, and there are
many reasons to accept that doctrine independently of the fact that content-external-
ism requires it.

Reasons to question CTM

I would now like to argue that CTM is erroneous and that the arguments for it involve
irreparable non-sequiturs. Let us start by putting forth one of the standard criticisms
of CTM – a criticism that is probably coeval with CTM itself.
Given any symbol-token S, and given any meaning (or referent) M, there is some
possible language L such that S means M in L. In English, the symbol-token “Socrates”
 Conceptual Atomism and the Computational Theory of Mind

refers to Socrates. But there are possible languages in which it refers to Plato and pos-
sible languages in which it doesn’t refer to anything. For exactly similar reasons, there
is a possible language in which the symbol-token “4” refers to Socrates, a possible lan-
guage in which it refers to Pluto, and so on.
Thus, if a computer transitions from “2+3=” to “5”, that is a valid transition only if it
is assumed those shapes instantiate English symbols. That transition is therefore valid only
from the viewpoint of somebody who speaks English, and its validity therefore presup-
poses the existence of thought. Genuinely symbolic activity, and thus computations, pre-
suppose thought, and therefore cannot be constitutive of it, at least not in all cases. (See
Dennett 1978: Chapter 6, Lycan 1984: 237, Kuczynski 2002, Searle 1984, 1992.11)
Here is a similar argument. Given any sequences of physical events, there is some
language L such that, in L, that sequence of events tokens a valid argument or a cor-
rectly executed mathematical operation. Indeed, given any sequence S of shapes, and
given anything that can be encoded in language – be it a poem, a political speech, or a
proposal for world peace – there is some language L such that S encodes that poem (or
political speech or…) in L. Given that not every single sequence of events is actually an
utterance of a beautiful poem (or political speech or…) it follows that no sequence of
events has any meaning at all merely in virtue of its shape. Nothing means anything
merely in virtue of its morphology. Convention alone can assign meaning to morphol-
ogy. There are conventions only where there are understandings and where there is
therefore thought. So there can be no assignment of meanings to morphology, and
thus no language, except where there is already thought. Thus it is viciously circular to
suppose that thought itself is linguistic.
What is the advocate of CTM to say in response to this point? One possible re-
sponse (indeed, one urged by Fodor himself) is this:
As you say, a token of the expression “2” doesn’t refer to the number two merely in
virtue of its shape. It is only because various symbolic conventions are at work that
an object with that shape refers to that number. But this has no relevance to CTM.
Mentalese symbols are not comparable to the symbols of a public language. Let 2
be a token of the Mentalese symbol that denotes the number two. It is not because
of convention that 2 refers to two,. The connection between “2” and two is medi-
ated by human thought. Hence the problem that you were discussing a moment
ago. But because the connection between 2 and two is not thus mediated, the
problem you bring up is irrelevant.

But then we have to ask: in virtue of what does 2 refer to two? Here is Fodor’s (1990, 1998)
answer: 2 refer to two in virtue of the fact that the 2 has a certain causal connection to
instances of the number two (i.e. to pairs of apples, pairs of shoes, and so on). Even if
Fodor is right about this, it doesn’t help CTM. As we’ve seen, the fact that 2 has certain
origins is causally and explanatorily inert. So if indeed it is in virtue of its causal origins

11. I believe that a version of this argument is found in Wittgenstein (1958) and (1983).
 Chapter 13.  Content-externalism and atomism 

that 2 refers to two, then from the viewpoint of the description and explanation of brain-
processes, 2 might as well be in the same category as a completely non-symbolic entity.
I would like to clarify this last point. Let M be some piece of paper that is also a
map. (So M might be the paper map that you have in the glove-compartment of your
car.) If you roll M up and strike Jones with it, various causal properties of M are hard
at work: M’s having such and such mass, and its moving with such and such velocity,
are responsible for the fact that Jones is now in pain. But M’s being a map hasn’t done
anything in this context. What we just said about maps is true of symbols. Let us sup-
pose that Fodor is right; let us suppose that each thought that a person has is a symbol-
token that is identical with, or realized by, some neural state. And let N be some neural
state that is also a token of Brown’s internal language. (So N is one of the neural states
making up Brown’s mind-brain.) Since neural states have causal powers, so does N.
But it doesn’t follow that N’s being a symbol has causal powers or, therefore, that N’s
being a symbol ever does anything. It could be that, in every context where N is doing
causal work, N’s being a symbol is as inert as M’s being a map was in the story told a
moment ago. (More precisely, it could be that, in every context where N’s having some
property or other – e.g. N’s having a certain morphology – is doing causal work, N’s
being a symbol is doing nothing.) But if anything having to do with Brown’s mind or
body is to be understood in terms of the fact that N is a symbol, then N’s being a sym-
bol cannot be causally impotent. N’s being a symbol must have effects and must, to that
extent, be comparable to N’s having a certain mass. If N is a symbol entirely in virtue
of its mode of origination, then N’s being a symbol does nothing. We’ve discussed at
length why, if x and y are exactly the same leaving aside their modes of origination, they
cannot differ in respect of their causal properties. (We also discussed the philosophical
confusions that have abetted a failure to see this.) If something is a symbol in virtue of
its mode of origination, i.e. if its being a symbol is identical with (or realized by) its
having a certain mode of origination, then a thing’s being a symbol does nothing. So if
our crania are populated by neural events that are also symbol-tokens, and if those
neural events qualify as symbol-tokens in virtue of their modes of origination, then
those neural events’ being symbol-tokens does nothing and is causally inert. In that
case, of course, nothing that happens in connection with our minds or bodies is to be
understood in terms of the fact that some of our neural states are symbol-tokens. So
the supposition that some of our neural states are symbolic is explanatorily idle, at least
to the extent that those neural states qualify as symbol-tokens by reason of their having
certain modes of origination.
So given a content-externalist conception of mental content, even if some of our
brain-states are symbol-tokens, their being symbol-tokens is causally idle, and CTM is
thus without explanatory power. But CTM fairs equally badly if content-externalism is
false. For, in that case, two brain-states couldn’t differ representationally unless they
differed in their intrinsic and, therefore, causally potent properties. So if content-ex-
ternalism is false, the representational properties of our brain states would be causally
potent, and the thought-processes realized by sequences of brain states would be con-
 Conceptual Atomism and the Computational Theory of Mind

tent-driven. It would immediately follow that CTM is wrong, since that doctrine de-
nies that thought-processes are content-driven.
Let us sum up. If the objector’s point is to prevail, semantics must be causally op-
erative. It must be in virtue of its referring to a certain number that 2 has the causal role
it does. But if semantics is operative, then CTM implodes, since the essence of that
doctrine is that thinking consists in sequences of symbol-tokenings that are not se-
mantics-driven.
Here is a different reason to question SCT and, therefore, CTM. As before, let 2 be
a token of the symbol in Brown’s internal language that denotes the number two. We
saw that, if CTM is to go through, the relationship between 2 and two cannot be con-
ventional. In fact, we saw something stronger: it must be wholly in virtue of its intrin-
sic properties – properties that no atom-for-atom duplicate of 2 lacks – that 2 refers to
two. So, if CTM is to go through, then 2 must intrinsically be a representation of the
number two; its being such a representation must be built right into it. But in that case,
it is hard to see why 2 would qualify as a symbol of the number of two. It seems that, on
the contrary, 2 would be a direct, non-symbolic apprehension of that number. This is a
point that we will develop in the upcoming section.
As Grice (1957) pointed out, the word “meaning” is ambiguous. CTM is plausibly
seen as involving a failure to see this. Smoke means fire. What does this mean? It means
that, given the presence of smoke, it may reasonably be inferred that there was recently
a fire. This kind of meaning is a strictly causal relation. (So this must be the kind of
meaning that Fodor has in mind when he says that it is entirely in virtue of its mode of
origination that 2 refers to the number two.) Further, instances of this kind of meaning
don’t presuppose the existence of thought. (So, to echo what we said a moment ago, this
must be the kind of meaning that Fodor has in mind when he says that linguistic mean-
ing doesn’t presuppose thought.) But when we say that smoke means fire, we don’t
mean that, where there is smoke, there is an awareness of fire. Meaning in this sense has
to do only with permissibility of inferences, and not with the actual making of inferences
or, more generally, with the actual occurrence of thoughts of any kind.
Of course, the word “means” sometimes expresses a psychologically pregnant notion.
If I say that “schnee ist weiss” means snow is white, what I am saying has implications as
regards people’s propositional attitudes, even though I am not directly making any state-
ment about those attitudes. For, as we’ve discussed, instances of that kind of meaning
presuppose the existence of conventions and, therefore, of understandings of some kind.
There is a third meaning of “meaning.” If I say “Bill means that you should become
a lawyer”, I am not making a statement about symbolic conventions, and I am explic-
itly making a statement about somebody’s thoughts. There obviously couldn’t be mean-
ing in this sense unless there were already thought. After all, for Bill to mean some-
thing by means of a certain noise is for a certain intention of Bill’s to cause, and
therefore pre-exist, that noise.
So far as CTM appears credible, it is because these various different meanings of
“meaning” are being collapsed. For CTM to work, meaning must be a strictly causal
 Chapter 13.  Content-externalism and atomism 

relation; further, instances of meaning mustn’t presuppose the existence of thought;

and, finally, any such instance must be a case of one thing’s being aware of another.
Given any one of these conditions, there is a disambiguation of the term “meaning”
that expresses a relation that satisfies that condition. But there is no one disambigua-
tion that expresses a relation that satisfies all three conditions.
Obviously Fodor is aware of the fact that “meaning” is ambiguous in the way just
described. So how are we to explain the mis-steps in Fodor’s argumentation? Here is my
guess as to how that question ought to be answered. If content-externalism were right,
the difference between somebody who had a concept of H2O and somebody who had a
concept of XYZ would lie solely in the fact that those people’s current conditions had
different geneses. In general, conceptual differences would be genetic differences, a cor-
ollary being that conception was to be understood in strictly genetic terms. In that case,
what concepts people had would be explained in terms of the modes of origination of
their current conditions. The relation of smoke to fire would not be fundamentally dif-
ferent from the relation that my concept of fire has to fire. Consequently, there would
be no fundamental difference between the psychologically antiseptic kind of meaning
involved in statements such as “smoke means fire” and the psychologically pregnant
kind involved in statements such as “what Smith means is that you should become a
lawyer.” So if content-externalism were correct, then the various kinds of meaning that
we just discussed would collapse into the strictly causal (“smoke means fire”) kind.
Fodor sees this. Fodor also accepts content-externalism. And that is why he would re-
ject our claim that CTM involves an equivocal use of the term “meaning.”
Fodor is right, I believe, to hold that if content-externalism is correct, then there is
ultimately but one kind of meaning – the strictly causal kind. But content-externalism
is not correct, as we saw in Part I. Thus, “meaning” is ambiguous, and CTM is guilty of
the equivocation described a moment ago.

Symbols as intermediaries: why CTM eviscerates the concept of a symbol

As we’ve seen, the connection between a Mentalese symbol and its meaning cannot be
conventional or otherwise mediated by thought. For CTM to work, 2 must naturally
mean two. (As before, 2 is a token of the symbol that, in somebody’s internal language,
denotes the number two.) In that case, 2 is just a direct awareness of that number. On
the one hand, we have a brain-state (consisting of some token of a Mentalese symbol);
on the other hand, we have a number; and the former is a direct awareness of the latter.
But it seems to be of the essence of symbolhood that a symbol is a kind of intermediary
(Kuczynski 2002, Horst 1996).
There is no symbolic awareness of Smith where there is a direct awareness of him. A
symbolic awareness of Smith consists in my encountering some symbol like “Smith”, and
interpreting it in the light of the relevant symbolic conventions. A direct awareness of
Smith would consist in my seeing him. The two kinds of awareness are antithetical to
 Conceptual Atomism and the Computational Theory of Mind

each other. (Of course, I could read and understand a sentence like “Smith is wise” while
also looking at Smith. But so far as I am doing the one, I am not doing the other.)
So unless we trivialize the notion of a symbol by saying that any awareness of any-
thing is ipso facto symbolic, it isn’t clear how Mentalese symbols (so-called) have any
significant similarity with actual symbols. This calls into question the viability of CTM,
since that doctrine just is the thesis that thinking consists in symbolic-manipulations
of a certain kind (Kuczynski 2002).
Of course, an advocate of CTM could defend against this by saying that the word
“symbol”, as used by CTM, is meant only in a metaphorical or otherwise extended sense.
(This, in fact, seems to be the position taken by Fodor 1975: 65–67, Lycan 1984: 237, and
other computationalists.) But in that case, CTM is reduced to the explanatorily innocu-
ous platitude that thinking consists in operations involving things that are like symbols,
in as much as both they and symbols represent things, but aren’t really symbols.

Why CTM is based on an equivocation on the term “form”

But underlying CTM is a confusion even more insidious than any of those just dis-
cussed. CTM identifies thinking with form-driven, or formal, manipulations of sym-
bols. But what reason is there to believe that a purely formal operation on symbols
could qualify as a cognitive achievement?
On the face of it, a formal operation is the antithesis of one that is thought-driven.
It is precisely when people don’t understand the formulas in front of them that they are
forced to rely on mechanical procedures. Where a neophyte needs to fall back on a
formal procedure to solve a problem (e.g. to derive an integral), a mathematician can
generate the right answer on the basis of genuine insight into the concepts involved.
The mathematician’s answer would seem to be more thought-driven than the neo-
phyte’s and, in general, it would seem that a symbolic operation is formal precisely to
the extent that it is not thought-driven.
Nonetheless, there is some basis for CTM’s identification of formal manipulations
of symbols with genuine cognitive achievements; and the grounds for that identifica-
tion can be distilled into the following argument:
Suppose that you use a truth-table to ascertain the truth-value of some sentence (or
sentence-schema). There is no denying that, in so doing, you have figured some-
thing out and thus deserve credit for some kind of cognitive achievement. And there
is no denying that, in this case, that cognitive achievement was generated through a
purely formal (or mechanical or algorithmic) operation on symbols.
So even though informal operations typically demand more cognitive where-
withal than formal ones, the latter do clearly constitute cognitive achievements and
are therefore cases of thinking. Thus, given the reasonable position that calculators
and computers successfully carry out formal symbolic manipulations, it readily fol-
 Chapter 13.  Content-externalism and atomism 

lows that they think. And, as we discussed, once it is granted that they can think,
both the tracking problem and the mind-body problem immediately vanish.

But this argument involves a fallacy. The words “form” and “formal” are ambiguous,
since they may refer to form in either the syntactic of the morphological sense, and the
argument just given equivocates between the different meanings of these terms. The
words “mechanical” and “automatic” are ambiguous in a comparable manner. The
whole of CTM involves a failure to register the different meanings of these various
expressions. And once these ambiguities are exposed, there remains little temptation
to advocate CTM.
(1) If Fred is wise, then Fred is a sentient being.
(2) If Fred is wise, then either Fred is wise or Fred is a cucumber.
Each of these is analytically true. But only the second is formally true. If someone tran-
sitions from “Fred is wise” to “Fred is sentient being”, he has made an analytically cor-
rect inference that is not formally valid. But if someone transitions from “Fred is wise”
to “either Fred is wise or Fred is a cucumber” he has made an inference that, in addi-
tion to being analytically correct, is also formally valid.
But what do these statements mean? What is the difference between “analytic” and
“formal” truth? What, in this context, do the words “formal” and “form” mean? Right
now I’m going to give the canonical answer to these questions – the answer that would
be given by mathematical logicians and also by advocates of CTM:
(SYN) (2) holds entirely in virtue of its syntactic structure. To know that (2) is correct,
it is necessary to know only its syntactic structure. It isn’t necessary to know
its semantics; it isn’t necessary to know, for example, what is meant by “wise”
or “cucumber.” By contrast, (1) does not hold merely in virtue of its syntactic
structure. There are syntactically identical sentences that express invalid in-
ferences, for example:
(3) If Fred is wise, then Fred is a rock.
To know that (1) is correct, one must know its semantics; one must know
what is meant by “wise” and “sentient being.”
So even though it is analytic, (1) is not formally correct, the reason being that
its syntax doesn’t guarantee its truth. By contrast, in addition to being analytic,
(2) is formally correct, the reason being its syntax does guarantee its truth.12
By way of anticipation, we will find that (SYN) is unsatisfactory in many ways. (In
particular, we will find that a knowledge of semantics is to some degree constitutive of
a knowledge of syntax and that (SYN) is therefore wrong, since it demands that knowl-
edge of syntax be entirely independent of knowledge of semantics.) But it is a good
point of departure, since it is approximately correct. Also, because it is the answer

12. Fodor (1981: 13-17, 1987: 18-20, 1990: 22, 1994: 294-298).
 Conceptual Atomism and the Computational Theory of Mind

given by advocates of CTM, we should grant it, if only for the sake of argument. If we
show that CTM fails given (SYN), then we will have shown that CTM fails relative to
its own assumptions, and is therefore uncontroversially false.
Supposing that (SYN) is correct, it is clear why CTM cannot be accepted. CTM is
false if it identifies thought-processes with syntax-driven operations, the reason being
that, as we will see, there can be such operations only in contexts there is already thought.
Aware of this, advocates of CTM identify thought-processes with morphology-driven
operations. But given only that a series of events is morphology-driven, that series
doesn’t qualify as a computation, i.e. it doesn’t qualify as a formal inference. Supposing
that our brain-processes are entirely morphology-driven, that is consistent with none of
them being computations.
Let us develop this argument. First of all, morphology is a property of expression-
tokens, not expression-types, since expression-types are abstract objects and thus don’t
have morphology. Two morphologically identical expression-tokens can have complete-
ly different syntactic forms, and two syntactically identical expression-tokens can have
two completely different morphologies.
Consider the symbol-token “snow is white.” There is a possible language L in
which that token, or at least one morphologically identical with it, means: if Beethoven
had been born in 1400, he would have invented the fugue. Considered as an expression
of L, that token has one syntactic structure. Considered as an expression of English, it
has an entirely different syntactic structure. (Incidentally, this shows that what syntac-
tic structure an expression has is a function of what it means – a point we will find to
be of inestimable importance.) In general, a sentence-token’s morphology, taken by
itself, imposes no constraints on its syntax.
Given any sentence-type, there is no limit to how different in respect of morphol-
ogy two tokens of it can be. Consider the sentence-type: “snow is white.” That sentence
could be tokened by sounds, hand-signal, light-rays, noises. And even within a single
one of these media, there is little limit to how different in respect of morphology two
tokens of that sentence can be. We could imagine a token of that sentence being ut-
tered by someone with a very deep voice over a period of a year; and we could imagine
a token of that same sentence being uttered by someone with a high voice (so high, in
fact, that it isn’t audible by humans) over a period of one nanosecond.
Of course, given two tokens of the sort just described, there will be a respect in
which they are isomorphic. But the isomorphism couldn’t be understood in strictly
acoustical terms, and would have to be understood, at least in part, in terms of the
meanings of the expressions involved. The correspondence could be established only
by showing that each of two sounds was an utterance of the word-type “snow”, and that
each of two other sounds was an utterance of the word-type “is”, and so on. Doing this
would involve much more than just pointing out the purely acoustical properties of the
sounds in question; it would involve discussing the psychological and social condi-
tions underlying the production of those sounds. The isomorphism in question would
thus not be of a strictly physical or morphological kind.
 Chapter 13.  Content-externalism and atomism 

Syntactic form has nothing to do with morphological form and, consequently, a

morphology-driven symbolic operation needn’t to any degree be syntax-driven. A be-
ing that is responding to form in the morphological sense is not, in virtue of that fact,
responding to form in the syntactic sense.

The concept of syntactic form

Here it becomes necessary to develop a point already hinted at. The syntax of an ex-
pression lies in its semantics. (2) has the same syntax as:
(4) If Smith is bald, then either Smith is bald or Smith is a bat.
(4)’s syntactic structure is a function of the meanings of the expressions of which it is
composed. Suppose that one of the occurrences of “Smith” were replaced with an occur-
rence of an expression belonging to a different semantic category. In that case, the result
would be meaninglessness. (Consider the pseudo-sentence that would result if “Smith”
were replaced with “cucumber.”) This shows that it is because “Smith” falls into a certain
semantic category that (4) is characterized by a certain syntactic structure. Obviously
(4)’s syntactic structure is determined, in part, by the fact that its main connective (“if…
then…”) has a certain meaning: that sentence would have an entirely different syntax if
the main connective were replaced with a different one (e.g. “and” or “or”), let alone with
an expression not belonging to the same semantic category as that connective.
Let us sum up. If you replace the occurrence of “Smith” in (4) with “Jones”, a sen-
tence with the same syntactic structure results. But nonsense results if you replace that
occurrence with an expression (e.g. “cucumber”, “elephant”, “not”) belonging to a com-
pletely different semantic category. A sentence’s syntax thus seems to lie in its more
abstract semantic properties. A syntactic description of a sentence is a low-resolution
description of its semantics. (Later we will replace this vague statement with a more
precise one.)
It follows that a manipulation of symbols is syntax-driven only if it is semantics-
driven. As we’ve seen, CTM cannot countenance the idea that the symbolic manipula-
tions constituting our cognitive processes are semantics-driven. Therefore, if CTM is
right, those manipulations are not form-driven in the syntactic sense of “form.” We’ve
already seen why, if CTM is right, those manipulations cannot be form-driven in the
morphological sense. Aside from syntactic and morphological form, it isn’t clear what
kind of form could possibly be relevant to CTM.
It thus appears that no disambiguation of the word “form” validates the idea that
our thinking consists in form-driven operations on symbols. So far as that idea has
credibility, it is because logical (or syntactical) form is being conflated with morpho-
logical form. This equivocation creates the illusion that purely morphology-driven
operations can miraculously have the property of realizing logic-driven, and therefore
cognitively pregnant, symbolic operations.
 Conceptual Atomism and the Computational Theory of Mind

Indexicality and the irreducibility of logical to syntactical form

For the sake of argument, let us set aside everything just said. Let us suppose, solely for
the sake of argument, that syntactic form is identical with morphological form or, at
least, that the two can be brought into some kind of meaningful alignment. That still
wouldn’t save CTM. This is because, given any language L that is rich enough to medi-
ate the thoughts that any person is capable of having, the syntactic structures of L’s
sentence-tokens will categorically fail to track the logical forms of the propositions
meant by those sentence-tokens.
Here I should take a moment to make a clarificatory point. Frege (1884) and Rus-
sell (1905) argued compellingly that logical form is very different from grammatical
form. Grammatically, “some man snores” has the same structure as “no man snores”,
and both have the same structure as “Smith snores.” But in terms of logical form, no
two of those sentences are comparable with each other. In general, where natural lan-
guages are concerned, grammatical and logical form pull apart.13
My point is that, given any language L, whether artificial or natural, that is rich enough
to express even a tiny fraction of the thoughts that any person can have, the logical forms
of L’s sentence-tokens will pull apart from their syntactic forms. There is no possibility of
the kind of isomorphism between syntactic and logical form that is needed by CTM.
Practically any thought that a person has is expressed by a sentence that has a con-
text-sensitive component. Consider the information that is coursing through your
psyche at this very moment. So far as it can be expressed by language, it is expressed by
sentences like “that is a hideous painting”, “I am feeling tired”, “now it is raining, but
yesterday it was not raining.” The italicized expressions are indexicals. If expressed in
language, the information that is useful to an organism’s survival is necessarily ex-
pressed by sentences that have an indexical component. (“That animal is running to-
wards me”, “I am now in terrible pain because this thorn is in my finger.”) The only
thoughts that can be expressed by indexical-free sentences are those that concern im-
mutable relations among concepts (e.g. “1+1=2”, “causes precede their effects”). Our
thoughts can be expressed by indexical-free sentences only in so far as they are of a
strictly mathematical or philosophical nature. Important as such thoughts are, they are
not very important to a creature’s survival or to its internal regulatory processes; and no
matter how intelligent a creature is, only a tiny fraction of its thoughts are of this sort.
Long ago, Strawson (1951) made it clear that, where there are indexicals, there is
necessarily a divergence between logical and syntactical form. (Kaplan (1977, 1989),

13. In any case, this is the philosophical conventional wisdom. It could be said that “Socrates
snores” and “some man snores” are grammatically different and that it is only relative to a kind
of “folk-grammar” – comparable to folk-physics or folk-medicine – that they have the same
grammatical form. I believe that this is the conceit underlying Montague’s (1974) work. This
general idea is validated by Chomsky’s work, since the latter shows how far “grammar” fails to
reveal the true structure of natural-language sentences. I discuss this in Footnote 17 of Kuc-
zynski (2006b).
 Chapter 13.  Content-externalism and atomism 

and also Barwise and Perry (1999), developed Strawson’s insight.) Let T and T* both
be utterances of “that is identical with that.” But suppose that in T both occurrences in
T of “that” refer to Fred, whereas in T* one occurrence refers to Fred while the other
refers to Bob. In that case, T encodes the logical truth Fred=Fred whereas T* encodes
the absurdity Fred=Bob. But, of course, T and T* have precisely the same syntactical
structure. By exactly analogous reasoning, a token of “if it is warm here, then it is warm
here” may encode an argument that is valid, invalid, or merely inductive (and thus
neither valid nor invalid), depending on whether both tokens of the indexical “here”
co-refer. It is easy to generate further illustrations of the principle at work here. For the
reasons stated in the previous paragraph, if our thoughts are mediated by linguistic
expressions, those expressions are replete with indexicals. As we’ve just seen, where
expressions of that kind are concerned, syntax cannot possibly track logical form.14 So
even if we waive all of the problems relating to the fact that morphology and syntax
have only a highly indirect and thus extremely unstable relationship with each other,
and even if we concede to CTM that syntax and morphology can be brought into some
kind of significant alignment, a form-driven manipulation of symbols is more likely
than not to fail to constitute a valid piece of reasoning.
Thus, even if we grant CTM all of its own assumptions – even if we concede that
there is a coincidence of morphology and syntax, and even if we concede that our
thoughts consist in form-driven operations on symbols – CTM fails to account for the
fact that our cognitive activity embodies a sensitivity to logical form.
What we’ve said can easily be condensed into a few points (it is easily seen that
some of these are logical consequences of others, and that the following list therefore
contains some redundancies):
(*) There is no morphological characterization of syntax.
(*) There is no syntactic characterization of logic.
(*) There is no morphological characterization of logic.
(*) Although a morphology-driven manipulation of symbols may in fact track
logical form, it does not track logical form in virtue of being morphology-
driven. (Although not directly argued for in this chapter, this last point is ob-
viously a consequence of the previous three.)

Analytic versus formal truth

We’ve seen that CTM involves a misunderstanding of the concept of “formal” truth
and that it fails to realize that the relevant kind of form is semantic in nature. Let us

14. This point is made clearly in Barwise and Perry (1999: 32-39) and Kaplan (1989: 508).
 Conceptual Atomism and the Computational Theory of Mind

now further examine the concept of formal truth. This will shed additional light on the
nature of the misunderstandings that underlie CTM.
We must start by distinguishing (once again) between analytic truth and formal
truth. “Either it is raining or it is not raining” is formally true (as well as analytically
true). By contrast, “anything wise is sentient” is not formally true, even though it is ana-
lytic. But what exactly is the difference between formal (logical) and analytic truth?
Sometimes a sentence is said to be “logically true” if it is true “under all reinterpreta-
tions of its non-logical constants” (Quine 1970: 50), where a “logical constant” is any of the
following expressions: “if…then…”, “and”, “or”, “not”, “=”, “some”, “all.” Sometimes modal
connectives, e.g. “possibly”, are put on this list. (Superficially, this definition seems to be
circular, since the term “logically true” is defined in terms of the expression “logical con-
stant”, making it seem that the term “logical” has been defined in terms of itself. But the
circularity vanishes if the expression “logical constant” is taken as an abbreviated way of
referring to the list containing all and only “not”, “all”, “=”, and so on.) Consider the follow-
ing four sentences:
(1) If Fred is wise, then Fred is a sentient being.
(2) If Fred is wise, then either Fred is wise or Fred is a cucumber.
(3) If Fred is wise, then Fred is a rock.
(4) If Smith is bald, then either Smith is bald or Smith is a bat.
(2) is true; so is (4) and so, indeed, is every other meaningful sentence that results
when the non-logical terms in (2) are uniformly replaced with others. For this reason,
(2) is formally true. (1) is true, and (3) is false. Thus, a false sentence can be generated
by uniformly replacing the non-logical expressions in (1) with other expressions. For
this reason, (1) is not a logical truth, even though it would traditionally be regarded as
an analytic or necessary truth (Quine 1970: 47–50).
There are two reasons why this analysis of logical truth is unsatisfactory. Suppose
that we defined the term “whole number” as “any number that is not a fraction or a real
number or a quaternion or an imaginary number…” Our definition would be exten-
sionally correct, but it would fail to identify the property that all and only whole num-
bers have in common (apart, of course, from the negative and disjunctive property of
not being either a negative number or a fraction or a real number…). Quine’s analysis
of logical truth is guilty of a similar defect, since it defines a logical truth as one that is
true under all substitutions of its component-expressions other than “and” and “or”
and “=” and… Even if that analysis is extensionally correct, it doesn’t say what it is for
a sentence to be logically true.
This brings us to the second problem with Quine’s analysis: it isn’t extensionally
correct. As Kaplan (1977, 1989) observed, sentence-tokens like “I am here now” are
naturally described as logically true. But the analysis of logical truth under examina-
tion cannot accommodate that fact. By the same token, that analysis cannot accom-
modate the fact that “I am not here now” is logically false.
 Chapter 13.  Content-externalism and atomism 

Can we say that a sentence is logically true iff it encodes an analytic proposition?
No. Given such a position, we couldn’t mark the obvious difference between
(KB) “any case of knowledge is a case of belief ”
and 4 by saying that the latter is, whereas the former is not, logically true. Nor would
that position accommodate the fact, mentioned a moment ago, that a token of “I am
here now” is logically true, at least on one reasonable delineation of the term “logically
true.” This is because, when I say “I am here now”, the proposition meant by my words
is identical, or at least equivalent, with the contingent proposition: JM is in Santa Bar-
bara on Oct. 6, 2006.

Formal truth as semantic truth

Given these points, I would propose a very different analysis of the concept of logical (or
formal) truth. Let T be some token of “I am here now.” Why is that token logically true,
even though it encodes an a posteriori, contingent proposition? Because it is a theorem of
the semantic rules of the language to which it belongs that it is true. (2) is logically true for
the same reason. In general, a sentence-token S, belonging to language L, is logically (or
formally) true exactly if S’s truth is a theorem of the semantic rules of L.
Consider the sentence:
(KB) “any case of knowledge is a case of belief ”
Given that (KB) means any case of knowledge is a case of belief, it logically follows that
(KB) is true. But the semantic rules of English do not themselves assign truth to that sen-
tence. Those rules merely assign a certain meaning to it. At that point, non-semantic (al-
beit purely conceptual and a priori) facts assign the truth-value true to it. English seman-
tics rules do not themselves assign the truth-value true to (KB) (or to any token thereof).
But a token T of “I am here now” is in a different category from (KB). The rules of
the language to which T belongs themselves assign truth to T. To see this, let us con-
sider those rules, leaving out irrelevant technicalities. One of those rules is “if some-
body x utters a token of FI have phiG, that token is true, i.e. it is assigned the truth-value
true, exactly if x has phi”. Another is: “if, in place x, there occurs a token of Fthis place
has phiG [or, equivalently, Fhere it is phiG], that token is assigned the truth-value true
exactly if x has phi.” Yet another of those semantic rules is: “if, at time x, there occurs a
token of Fthis time has phiG [or, equivalently, Fnow it is phiG], that token is assigned the
truth-value true exactly if x has phi.” Given these and a few other strictly semantic
facts, it follows that it is a theorem of English semantics that any token of “I am here
now” is assigned truth. That is why any such utterance is logically true, even though
what such an utterance semantically encodes is contingent and a posteriori.
 Conceptual Atomism and the Computational Theory of Mind

Another example may be appropriate. Any token T * of:

(RNR) “either it is raining or it is not the case that it is raining”
is syntactically (or formally) true. To see why this is so, let us consider the semantic rules
that assign meaning to such a token. (As before, we will set aside irrelevant technicali-
ties.) One of those rules is: “a token of Fit is not the case that SG is true, i.e. is assigned the
truth-value true, iff S is false.” Another is: “a token of Feither S or QG is assigned the truth-
value true iff either S is assigned the truth-value true or Q is assigned that truth-value.”
Given these facts, and a few others of a strictly semantic nature, it follows that it is a
theorem of English-semantics that T* is assigned the truth-value true. So T and T* are
logically true because of how they are assigned the meanings that they in fact have.
This suggests that syntax is an aspect of meaning, and that the concept of syntactic
(“formal”) truth is not a meaning-innocent one. Given what we’ve seen, a sentence-
token’s syntax would seem to lie in how it means what it means (as opposed to what it
means). In a moment, we will substantiate this thesis as to the nature of syntax. We
should point out that, if that thesis is correct, then T and T* are logically true because
of how they mean what they mean, and their being formally true is to be understood
in terms of their semantics – or, more accurately, in terms of their pre-semantics (i.e.
in terms of the semantic rules that assign them their semantics).
It is often asked “in what sense are logical truths formally true?” (Here the term
“logical truth” is to be taken to refer to the kinds of sentences studied by mathematical
logicians. So in this context “if everything is green, then it is not the case that nothing
is not not green” is a “logical truth”, as is “either it is snowing or it is not the case that it
is snowing”, whereas “there can be pain only if there is sentience” is not a logical truth.)
As we’ve seen, a sentence S, belonging to language L, is logically true just in case it is a
theorem of L’s semantic rules that S is true. Thus, a sentence is logically true just in case
its derivation-tree is a proof of it. So a sentence is logically true just in case it holds in
virtue of how it is assigned meaning that it has. Two sentences (belonging to some one
language) that have different meanings may coincide in respect of how they are as-
signed those meanings by the semantic rules of the language to which they belong. So
a sentence S is logically true in virtue of having properties that it can have in common
with sentences that have meanings completely different from S and that resemble S
only in respect of the structures of their derivation-trees. Thus, if S is logically true, it
is true in virtue of its having some abstract structural – or, as we might also put it,
formal – property. A sentence is formally true if the semantic rules that assign it mean-
ing do so in a certain way. Formal truth therefore has nothing do with morphology. It
is an abstract property of semantic rules. Formal truth is a property of a sentence’s (or,
strictly speaking, a sentence-token’s) pre-semantics. “Either it is snowing or it is not
the case that it is snowing” has the same form, in the relevant sense of the word, as
“either it is raining or it is not the case that it is raining.” The pre-semantics of the one
sentence is, of course, different from the pre-semantics of the other. (The pre-seman-
tics of the one comprises a rule that assigns meaning to the word “snow”, whereas the
 Chapter 13.  Content-externalism and atomism 

pre-semantics of the other does not comprise such a rule.) Nonetheless, the pre-se-
mantics of the one sentence is obviously structurally similar to that of the other, that
being why those sentences are “formally” identical. Any sentence having that abstract,
formal property is true. (No sentence of the form either P or it is not the case that P is
false.) So any sentence having that property is ipso facto true, i.e. any such sentence is
true in virtue of that fact. Thus “either it is snowing or it is not the case that it is snow-
ing” is true in virtue of its “form” – meaning that any sentence with a structurally
similar pre-semantics is also true. So in what sense is logical truth formal truth? Logi-
cal truth is truth in virtue of structural, and thus formal, facts about pre-semantics.

The relevance of these points to CTM

Formal truth is semantic truth. Indeed, there is no notion that is more semantic than
that of formal truth, since E is a formally true expression just in case E’s truth is a theo-
rem of the semantic rules that assign it meaning. It is therefore clear that unless a crea-
ture is responding to the semantic properties of the symbols that it houses, it is not re-
sponding to the forms of those symbols, at least not in any relevant sense of the word.
Our analysis is also consistent with the close connection between the concepts of
formal truth and computation. A statement is formally true or formally false exactly if
its truth-value can be computed on the basis of the rules that assign it meaning.
One might make the following objection to this analysis:
You may be right to say that all cases of formal truth are cases of semantic truth.
But the converse doesn’t hold. Given the semantic rules assigning meaning to “1”,
“2”, “+”, and so on, it is a semantic theorem that “12+13=25.” But surely that is not
a truth of logic. Its truth is indeed a theorem of the relevant semantic conventions,
and the proposition meant by it is true on purely conceptual grounds. But that
proposition is not sufficiently generic. Its subject-matter is far too specialized to
grant it membership in the class of logical truths.15 A sentence expresses a logical
truth only if the truth of that sentence turns on facts about highly general con-
cepts – concepts like disjunction, negation, and the like. Concepts like 12, 25, and
addition are much too specialized.

Even if this objection is correct, it concedes that semantic truth is a necessary condition
for formal truth, and that objection thus doesn’t help CTM at all.16

15. See Russell (1903: Chapter I) for a similar point.

16. Also, that objection isn’t correct. If it is a theorem of the semantic rules that assign it mea-
ning that “12+13=26” is true, then that sentence is a truth of logic. (Logicism – the view that
arithmetical truth is formal truth – failed precisely because it turned out not to be possible to
identify a language L such that all and only the theorems of arithmetic were semantic theorems
of L.) Also, logical truth cannot be identified with general truth. Even though it isn’t general, “if
Smith is here, then Smith is not not here” is logically true. Even though it is more general than
 Conceptual Atomism and the Computational Theory of Mind

We’ve seen that the word “form” is ambiguous, and that CTM involves a failure to
register this ambiguity. Points similar to those just made in connection with the word
“form”, and in connection with CTM’s use of it, are true of a number of other expres-
sions that occur in arguments on behalf of CTM. Among these other expressions are
“automatic procedure”, “mechanical procedure”, and “algorithm.”
It is often said that the truth-values of some sentences can be determined by means
of “mechanical procedures.” This claim is ambiguous, and neither disambiguation fa-
vors CTM. Consider the following four sentences:
(1) Either Smith is tall or Smith is not tall.
(2) If Smith is tall, then it is not the case that Smith is not tall.
(3) There are continuous functions that cannot be differentiated at any point.
(4) It is false to say that laws are simply commands and do not necessarily differ
in any significant respect from a gunman’s threats.
No one would say that there was any mechanical way of determining whether (3) was
true. In fact, (3) is true. But that was established on the basis of creative insight, not on
the basis of an application of a mechanical test. The same point holds of (4), supposing
that (as H.L.A. Hart argues17) it is in fact true. But there is a strictly mechanical proce-
dure determining whether (1) and (2) are true or not. We need only apply the method
of truth-tables.
In this context, what does the term “mechanical procedure” mean? The naïve an-
swer is to say that a mechanical procedure is one that a machine could implement. But
a machine couldn’t determine whether (1) and (2) were true or not. Obviously there is
no reason why we couldn’t construct a machine that scanned (1) and (2), and then
wrote “true” next to (1) and “false” next to (2), and did the same for any other sentence
whose main connective was a truth-functional operator like “and” or “either…or…”.
But, as we saw earlier, unless the machine’s behavior is semantics-driven, it isn’t figur-
ing out whether those sentences are true or false.
If we allow that the behavior of a “machine” can be semantics-driven, then a ma-
chine is no less capable than we are of figuring out whether (3) and (4) are true. In that
case, if we say that a mechanical procedure is one that a “machine” can apply, we are
falsely saying that there is a mechanical procedure for determining whether (3) and (4)
are true and, by the same token, for writing effective literary criticism and composing

the sentence just considered, “Smith cannot possibly occupy a given place at a given time and
also fail to occupy that place at that same time” is not a logical truth (even though it is analytic).
And even though it is maximally general, “nothing can pre-exist itself in a non-cyclical time-
line” is not logically true (even though it is analytic). It is because it is semantically true that the
first sentence is a truth of logic, and it is because they are not semantically true that the second
and third sentences are not truths of logic. Any overlap between the class of general truths and
logical truths is derivative of the overlap between semantic truths and general truths.
17. Hart (1961).
 Chapter 13.  Content-externalism and atomism 

great symphonies. So it is either false to say that a mechanical procedure is one that a
machine can implement or it is true only on an irrelevant disambiguation.
Given the points made a moment ago concerning syntax, I would propose a differ-
ent answer to the question: what is a mechanical procedure? It is a semantic theorem that
(1) and (2) are true. It is not such a theorem that (3) and (4) are true (even though they
are both conceptually true). As we noted, it is generally held that there is a “mechanical
procedure” for ascertaining the truth-values of (1) and (2), but not of (3) and (4). This
suggests that there is a “mechanical procedure” for determining the truth-value of a sen-
tence (or sentence-token) S, belonging to language L, exactly if it is a theorem of the se-
mantic rules of L that S is true. If this is right, then the execution of a “mechanical” pro-
cedure presupposes a heavy backlog of semantic and extra-semantic knowledge.

Syntax as meaning-how

We’ve already seen some reason to believe that CTM may be abetted by a misunderstand-
ing of the term syntax. For we’ve seen that the syntactic properties of a sentence are among
its semantic properties, even though the principle arguments for CTM assume the exact
opposite. But what exactly is syntax, and what is it for two sentences to have the same syn-
tax? Given plausible answers to these questions, it becomes even more apparent than before
that CTM involves a misuse of the concept of syntax.
Let us start with some truisms. Two sentence-tokens have the same semantics if
they mean the same thing; semantics is meaning. But there are different ways that a
given semantics can be assigned to a noise or ink-mark. It would seem that
(BKC) “Brutus killed Caesar”
encodes the same proposition as
(CKB) “Caesar was killed by Brutus.”
Some would deny this, saying that (BKC) and (CKB) encode distinct but equivalent
propositions. But it seems more reasonable to suppose that those inscriptions differ
not in respect of the proposition that they encode, but in respect of the grammatical
mechanisms that link them to some one proposition.
The proposition meant by “triangles have three sides” is logically equivalent with
that meant by “1+1=2.” But the relationship between these two propositions obviously
isn’t comparable to relationship between the proposition meant by (CKB) and that
meant by (BKC). In the one case but not the other, it can plausibly be said that we are
dealing only with “purely verbal” differences. A reasonable synthesis of these facts sis
to be found in the supposition that (CKB) and (BKC) differ not in respect of what they
mean, but in respect of how they mean what they mean.
 Conceptual Atomism and the Computational Theory of Mind

It is uncontroversial that, from a linguist’s standpoint, (CKB) and (BKC) are syn-
tactically different.18 Given that those tokens encode the same proposition, this sug-
gests that those tokens are syntactically different because of how they mean what they
mean. In general, syntax is meaning-how, whereas semantics is meaning-what.
(BKC) is syntactically identical with:
(JKS) “Jones kicked Smith.”
Why is this? (JKS) and (BKC) have structurally identical “derivation-trees.” Though they
have different meanings, they are “hooked up” to those meanings in similar ways.
Since the syntax of an expression lies in the relationship that it bears to its mean-
ing, it is absurd to say that syntax-driven operations are semantically innocent.
Independently of CTM and of our criticisms of it, it is clear that, even though they
have different propositions for their meanings, it is in virtue of some kind of meaning-
involving similarity that (BKC) and
(SSJ) “Smith saw Jones”
are syntactically identical. Each of “Smith” and “Brutus” is a singular term; each of
“saw” and “killed” are past-tense transitive verbs; and so on. These are semantic facts.
A singular term is one that refers to a certain kind of object; a verb is a term that ex-
presses a certain kind of relation; and so on. It is obviously in virtue of these semantic
facts that (SSJ) and (BKC) have the same syntax. Facts about syntax are facts about
meaning, even though two expressions can have the same meaning and different syn-
taxes and vice versa. The tension between these two facts vanishes when it is seen that
an expression’s syntax lies not in what it means, but in how it means what it means.
Let
(*) XYZ
be some sentence of Finnish that means:
(**) if aardvarks were composers, then given their psychologies, they would prob-
ably be inclined to express their musicality in the form of highly contrapuntal
but melodically impoverished compositions.
For the sake of argument, let us suppose that you don’t speak Finnish. If a Finn tells
you that (*) has (**) for its meaning, there is a sense in which you now know the mean-
ing of (*), but there is also a sense in which you don’t. A Finn does, whereas you do not,
have an understanding of the semantic decomposition, to use Jerrold Katz’s (1972)
expression, of XYZ. You don’t know which part of XYZ means aardvarks, or which
part means musicality, and so on. But even if you have a dictionary, and thus know
which words in (*) are the translations of “melodically”, “aardvarks”, and so on, you still

18. See, for example, McCawley (1998).

 Chapter 13.  Content-externalism and atomism 

won’t know why those words are ordered or inflected in the way that they are, so far as
you are even able to distinguish inflections from word-roots.
So there is at least one reasonable delineation of the term “meaning” relative to
which a Finn does, whereas you do not, know the meaning of XYZ, notwithstanding
that you know which proposition XYZ encodes. What a Finn knows that you do not is
the derivation-tree of XYZ. This follows directly from our description of the situation.
It is consistent with existing usage of the term “syntax” to say that a Finn does, where-
as you do not, know XYZ’s syntax.
It follows that the concept of syntax is not a meaning-innocent notion and that CTM
is false so far as it assumes otherwise. A corollary is that, so far as the term “mechanical
procedure” has syntactic content, a sensitivity to meaning is a prerequisite to executing
such a procedure. For reasons already considered, this is problematic for CTM.
Let us consider some possible responses to our analysis:
In natural language, grammatical and logical form diverge, as you’ve discussed.
“Socrates is wise” and “no one is wise” have the same grammatical form, even
though they have very different logical forms. But we can construct languages in
which the logical differences between “Socrates is bald” and “no one is bald” are
not obscured by grammatical similarities and in which, in general, grammatical
and logical form coincide.
The word “syntax” can refer either to the syntax of one of these idealized lan-
guages or it can refer to the syntax of natural languages. Linguists typically use
that word in the second sense, and logicians typically use it in the first.
So far as you have shown anything, it is only that, in the linguist’s sense of the
word “syntax”, sensitivity to meaning is a precondition for engaging in syntax-
driven operations. You have not shown that anything comparable holds in con-
nection with the other disambiguation of that word. Since it is the latter disam-
biguation that matters in this context, you have not shown anything of
significance.

For the sake of argument, let us grant the objector’s supposition that syntactic and
logical form coincide where Mentalese expressions are concerned. That reinforces
more than it undermines our analysis of syntax and our related criticism of CTM.
Let P and P* be the propositions meant by “nothing is wise” and “Socrates is wise”,
respectively. The grammatical mechanisms that hook up the former with P are similar
to those that hook up the latter with P*. More precisely, the derivation-tree for the one
sentence is structurally the same as the derivation-tree for the other. That is what cre-
ates the illusion of a semantic parallelism, i.e. of a parallelism in respect of the proposi-
tions that those sentences express.
Now let us suppose that English* is just like English except that, in English*, logi-
cal differences are always marked by syntactic differences. So in English* one doesn’t
say “nothing is wise”. Rather, one produces a sentence whose grammatical form match-
es its logical form, e.g. “the property of being wise is uninstantiated.”
 Conceptual Atomism and the Computational Theory of Mind

In English*, syntax and logical form coalesce because dissimilarities between what
sentences mean are not covered up by similarities in respect of how they have these
meanings, and similarities between what sentences mean are not covered up by dis-
similarities in respect of how they have those meanings.
For exactly similar reasons, if Mentalese is logically perspicuous, that is because
(dis)similarities between what two expressions mean correspond to (dis)similarities in
respect of how they have those meanings. So contrary to what the objector says, syntax
is no less a meaning-involving affair than where Mentalese is concerned than where
English is concerned.19
One of the objectives of the present work has been to show that facts about pre-
semantics are often mistaken for semantics itself. The how of meaning is displaced
onto its what. Given a failure to distinguish meaning-what from meaning-how, it is no
wonder that what should in fact be described as “syntax” ends up being described as
“semantics.” It is because of this muddle that there is nowhere for “syntax” to go other
than into the domain of pure morphology.
It is therefore not unreasonable to conjecture that a failure to distinguish seman-
tics from pre-semantics is responsible for CTM’s presupposition that syntax is identi-
cal, or at least capable of being meaningfully aligned with, pure morphology. In its
turn, the idea that morphology by itself has some kind of significant connection with
syntax validates the idea that, from at least one point of view, a symbol-token just is a
bit of morphology.

Another disambiguation of the term “syntax”

A linguist would say that

(i) “snow is white or snow is not white”

19. Let us consider another objection to our analysis:

“You say that a sentence’s syntax lies not in what it means, but in how it means it. It seems a
truism that a sentence’s semantics lies in what it means. So your position is that a sentence’s
syntax lies not in what its semantics is, but in how it has that semantics. But earlier you said that
a sentence’s syntax is an aspect of its semantics. So you have both affirmed and denied that syn-
tax is an aspect of semantics.”
We’ve seen that a sentence’s syntax is bound up with its semantics: the former lies in how that
sentence has the latter. So syntax is a meaning-involving affair. We can take my earlier statements
to the effect that “syntax is an aspect of semantics” as an abbreviated way of saying this. It is readily
seen that this will not have any negative effects on the arguments in which that statement figured,
since those arguments depended on the thesis that syntax was an aspect of meaning – it being left
open whether that aspect concerned the how, as opposed to the what, of meaning. So while I
concede that, technically speaking, I have used the term “semantics” equivocally, the equivocation
has no bearing on any of our arguments.
 Chapter 13.  Content-externalism and atomism 

has the same syntactic structure as

(ii) “snow is white and snow is not white.”
But a mathematical logician would not say this. This isn’t because the mathematical
logician disagrees with the linguist; it is only because the two use the word “syntax” in
different ways.
In any case, it is easily seen that the mathematical logician’s disambiguation of that
term is no more capable than the linguist’s of validating CTM. For the mathematical
logician, two sentences have the same syntax iff there are no significant proof-theo-
retic differences between them.20 (i) and (ii) are significantly different from that view-
point. So are (ii) and
(iii) “if Smith is green, then either Smith is green or he is not green”,
even though both (i) and (iii) are both “syntactically true” and are therefore both prov-
able. Two syntactically true sentences are syntactically identical if they are analogous
in respect of their proof-theoretic properties, the same thing mutatis mutandis holding
of syntactically false sentences. (In the last sentence, I used the word “analogous”, and
not “identical”, because two sentences that are, from the logician’s viewpoint, syntacti-
cally identical may obviously have different proof-theoretic properties. “Either snow is
white or it is not the case that snow is white” has different proof-theoretic properties
from “either grass is green or it is not the case that grass is green”, even though those
sentences have the same syntax. Clearly, those sentences have the same syntax because
they have parallel, not identical, proof-theoretic properties.)
Of course, not all sentences are syntactically true (or syntactically false). But be-
cause we know what it is for two syntactically true sentences to be syntactically identi-
cal, it is easy to produce a general criterion of syntactic identity – one that applies to
any pair of sentences, regardless of whether its members are syntactically true or not.
Any sentence, whether syntactically true or not, can be embedded in a complex sen-
tence that is syntactically true. The sentence “if Smith is tall, then something is tall” is
syntactically true, even though “Smith is tall” is not. In light of this, let S and S* be any
two sentences, whether syntactically true or not; and let C(S) be any syntactically-true
sentence of the form…S…If C(S) is syntactically identical with C(S*) (i.e. with …S*…),
then S and S* are syntactically identical; and if not, not.
When the mathematical logician talks about “syntax”, he is talking about an aspect
of semantics. If he says that a sentence is syntactically true, he is saying that it is a theo-
rem of the rules that assign it meaning that it is true; and if he says that two sentences S
and S* are syntactically identical, he is saying that the proof of the theorem that C(S) is

20. Of course, what counts as a “significant” proof-theoretic difference is a function of the dia-
lectical context. This suggests that it probably isn’t possible to speak of the concept of syntactic
structure, even if we confine ourselves to the logician’s disambiguation of the word “syntax.” Si-
milar remarks hold in connection with the linguist’s disambiguation of that term.
 Conceptual Atomism and the Computational Theory of Mind

true is relevantly similar to the proof of the corresponding theorem for C(S*), where C(S)
and C(S*) are defined as before. Like the linguist, the mathematical logician uses the
term “syntax” to refer to meaning-how. The logician and the linguist operate with differ-
ent contextualizations of the notion of meaning-how. But that doesn’t matter in this con-
text. What matters is that, when either the mathematical logician or the linguist uses it,
the term “syntax” refers to an aspect of meaning and, consequently, that a syntax-driven
operation is meaning-driven. We may conclude that neither the linguist’s nor the math-
ematical logician’s disambiguation of the term “syntax” favors CTM.21

More on the concept of a mechanical procedure

There is another delineation of the “mechanical procedure.” Let us see if it does a better
job of validating CTM than the delineation just considered.
Let LM be the system of semantic rules that assigns the number one to “1”, the
number two to “two”, the operation of addition to “+”, and so on. As almost everyone
past the age of six knows, there are special techniques involving those rules that enable
one to add (or multiply or divide…) large numbers. If one wants to add 1,297,967 and

21. Consider the following two scenarios. First scenario: nations A and B are engaged in a dis-
pute. Jones, who is the dictator of nation A, responds by waging an all-out attack on B. The attack
involves only foot-soldiers: no planes or long-range missiles are used. Second scenario: nations
C and D are engaged in a dispute. Smith, who is the dictator of C, responds by waging on all-out
attack on D. But Smith, unlike Jones, uses only long-range missiles: no foot-soldiers are invol-
ved. The question “how did Jones resolve the dispute between his nation and B?” may or may not
receive the same answer as the question “how did Smith resolve the dispute between his nation
and D?” A military strategist might say that those questions are to be given different answers, the
reason being that, from his viewpoint, Smith and Jones behave in very different ways. (From a
military strategist’s viewpoint, there is an important difference between using foot-soldiers and
using long-range missiles.) But a psychologist might say that those questions are to be given the
same answer, the reason being that, in every characterologically significant respect, Smith and
Jones behaved in the same way. (Both used brute force instead of diplomacy or economic measu-
res. The fact that one chose to use long-range missiles, while the other chose to use foot-soldiers,
doesn’t indicate any fundamental difference in their characters.)
Both linguists and mathematical logicians agree that syntax is meaning-how. But the ques-
tion “how is ‘snow is white or snow is not white’ assigned the meaning that it has?” may or may
not be given the same answer as the question “how is ‘snow is white and snow is not white’ assi-
gned the meaning that it has?” The linguist is interested in issues relating to learnability and
mental representation; and from that viewpoint, these two questions are to be given the same
answer. The mathematician is interested in issues relating to provability; and from that view-
point, those questions are to be given different answers. The linguist and the mathematical logi-
cian agree that syntax is meaning-how, even though, given the differences in their concerns, they
operate with different contextualizations of that notion. Thus, a case can b made that the term
“syntax” is not, ultimately, ambiguous.
 Chapter 13.  Content-externalism and atomism 

7,645, one writes down the corresponding numerals, making sure to line up with the
“5” with the “7”, the “4” with the “6”, and so on. Once that is done, adding those two
numbers is reduced to adding a series of single digit numbers. So what would other-
wise be a task requiring Herculean mathematical prowess becomes a task that is en-
tirely “mechanical”, meaning that no intelligence (over and above such as is possessed
by anyone past infancy who is of normal cognitive ability) is required to execute it.
Obviously this delineation of the term “mechanical” doesn’t validate CTM. In or-
der to implement the procedure just described, one must have a formidable repertoire
of perceptual and cognitive capabilities. One must be able to see, hear, and write down
symbols; one must have mastered a system of symbolic conventions. Only a creature
that already has considerable cognitive powers can engage in operations that are “me-
chanical” in the sense just described. It would therefore be absurd to suppose that the
foundations of cognitive life consisted in the implementation of such operations.22
There is a third delineation of the term “mechanical procedure.” Given the exis-
tence of certain symbolic conventions (e.g. the convention that “1” denotes the num-
ber one), there is a function that assigns truth-values to arithmetical statements on the
basis of their shapes. (In any case, this is what CTM assumes, since that doctrine as-
sumes that there is, or at least can be, an extremely strict correspondence between se-
mantics and morphology. We will see in the next chapter that this statement is only
approximately true and that the needed qualifications expose grave problems with
CTM. But, for argument’s sake, let us now grant this assumption of CTM’s.) There is a
function that assigns truth to “1+1=2” on the basis of its shape and that assigns falsity
to “1+1=3” on the basis of its shape. And there is no difficulty constructing a machine
M such that, given the left-side of an arithmetical equation, M produces a shape that is
assigned truth by the function in question. (So given “12+12=”, M writes down, or
otherwise generates, a “24”, thus producing a shape – “12+12=24” – that is assigned
truth by the function in question.) The activity of a machine like M is obviously me-
chanical in a quite strict sense of the word: it is purely physics-driven; consideration of
meaning plays no part.
One is tempted to say of machines like M that they add, subtract, and so on. Of
course, we add and subtract; and, in so doing, we are thinking. Given this, it seems to
follow straightforwardly that, in virtue of engaging in purely mechanical procedures,
machines like M are thinking. In general, purely mechanical activity can yield thought.
But such a position involves a blunder that we’ve already discussed. Given any
inscription, no matter what its morphology, there are infinitely many languages L such
that, in L, that inscription means nothing or means something false. Since M’s behav-
ior isn’t driven by a sensitivity to the relevant semantic conventions, that behavior does
not, by itself, constitute a case of adding twelve to itself.

22. An exceptionally clear and powerful defense of this general line of thought is found in
Horst (1996). See Wittgenstein (1974) and Coulter (1983) for similar viewpoints.
 Conceptual Atomism and the Computational Theory of Mind

This must be understood aright. Given that M has written “12+12=24”, there has
been a genuine cognitive achievement: there has been a case of twelve being correctly
added to itself. But the structure of that achievement is very different from that as-
cribed to it by the position described in the last paragraphs.
This can be made clear through a story. Smith is somebody who knows about the
conventions constitutive of L. Also, Smith knows that M has been so constructed that
it produces inscriptions that, relative to those conventions, constitute true arithmetical
statements. Finally, Smith punches in “12+12=”, and watches as M generates a “24.” (To
simplify discussion, let us suppose that Smith is so mathematically incompetent that
he can’t add twelve to itself without the assistance of a calculator.)
Twelve has been correctly added to itself. But that achievement consists not in
anything that the calculator does, but in Smith’s being able to put the calculator’s activ-
ity into the appropriate socio-psychological context. Smith realizes that, given what
the conventions of L are and given what M is the fact that M produces a token of “24”
under the circumstances described warrants the conclusion that 12+12=24. A purely
mechanical procedure (M’s engaging in certain strictly physics-driven manipulation of
shapes) is constitutively, and not merely causally, involved in Smith’s having this recog-
nition. This is shown by the fact that, if M malfunctioned, but by some coincidence
produced a “24”, Smith wouldn’t have acquired any mathematical knowledge. (Suppos-
ing that Smith had every reason to believe that the machine wouldn’t malfunction, we’d
have a case of justified true belief that wasn’t knowledge.) But that physics-driven pro-
cess by itself doesn’t constitute such an achievement. It is Smith’s having a certain rec-
ognition that does so.23
A procedure that is “mechanical”, in the sense just described, doesn’t presuppose
any mental faculties of any kind. But, as we just saw, such a procedure also fails to gen-
erate anything mental. So even though it wouldn’t be viciously circular to suppose that
our thinking is realized by mechanical processes of the kind just described, such a sup-
position would still be quite erroneous. This third delineation of the term “mechanical
procedure” therefore fails to validate CTM.
An “algorithm” is a “mechanical procedure” for deciding the truth-values of state-
ments. Given what we’ve just said, it follows there is no delineation of the expression
“algorithm” that validates CTM.

Linguistic communication as involving isomorphism, not identity,

of thought (revisited)

Earlier we discussed why linguistic communication demands that speaker and auditor
have isomorphic, not coincident, thoughts. Before that, we discussed why what a sen-
tence communicates consists less in what is literally meant by it than in what the auditor

23. See Wittgenstein (1974), Coulter (1983), Horst (1996).

 Chapter 13.  Content-externalism and atomism 

learns in the process of figuring out what is literally meant by it. In this section, we’ve
seen why “formal” truth is syntactical truth and why, consequently, a sentence-token is
“formally true” in virtue of how it means what it means. A short story will help show how
these points relate to one another, and also why they have so often been overlooked.
Aaronson is a scientist who has generated a great deal of experimental data. He
sends a report of that data to me. His report contains no analysis – just raw data. My
job is to distil that data into a theoretically significant statement. In the process of do-
ing so, I must add two 87-digit numbers. Even though I don’t have a calculator, I am
able to compute the relevant sum, thanks to the algorithms that I learned as a third-
grader. It is beyond my powers to figure out to what exact number an 87-digit number-
expressions refers. So the identities of the two numbers that I am adding are unknown
to me, as is the identity of their sum.
I include my computations in my lab-report, which I then send to Brown. Brown
is no more able than I am to excogitate the reference of an 87-digit number-expression.
But, for reasons similar to those just discussed, he can still use my written record of my
computation to analyze what I have sent him. (Brown’s job is to mediate between sci-
entists and non-scientifically inclined policy-makers. So his job is to convert my anal-
ysis into a form that is meaningful in the context of a policy-discussion.) Brown suc-
cessfully condenses my analysis into a statement that contains only short, easily
understood number expressions. He then relays that analysis to policy-maker Green.
Here information was transmitted from Aaronson to Green. Many of the symbols
that were instrumental to that accomplishment were not understood. But those sym-
bol-tokens were not therefore useless. Even though nobody fully understood the 87-
digit number-expressions just described, those expressions facilitated the flow of infor-
mation from person to person. We thus see corroboration for our earlier point that part
of what makes language so useful is precisely that it isn’t in lock-step with thought.
But one might argue that this digression ultimately undermines our position.
Those number-expressions were useful in this context, despite their not being fully
understood, only because there are formal rules for operating with them. So even
though it may support our views concerning the distinction between literal and cogni-
tive meaning, this little digression undermines our critique of CTM and is therefore,
on balance, to the discredit of our position.
This is not so. What does it mean to say that, in the story just told, my operations with
those number-expressions were “formal”? I knew the rules assigning meanings (referents)
to those digits; I knew that “1” denotes one; that juxtaposing single-digit numerals m and
n denotes the result of adding the number denoted by n to the number that results when
the number denoted by m is multiplied by ten; that “+” denotes the operation of addition;
and so on. My having (so-called) formal knowledge thus consists in my knowing how
those 87-digit expressions meant what they mean without knowing what they mean, and
so far as I am performing any (so-called) formal operations, it is because my behavior is
driven by such knowledge-how. Operations that are formal in this sense obviously cannot
 Conceptual Atomism and the Computational Theory of Mind

occur except where there is already semantic knowledge, and this disambiguation of the
term “formal operation” thus doesn’t validate CTM.
Let us pause to make a point that doesn’t have any special connection to CTM. In
Part I, we argued at length that much of the information that person A takes away from
a verbal exchange with person B is absorbed in the process of the A’s figuring out what
is meant by B’s words. What is actually meant by those words is, of course, important.
But its importance lies not in its coinciding with what is transmitted, but in its direct-
ing the auditor’s information-gathering efforts in a certain direction.
The story just told corroborates this. Nobody in our story knew what the 87-digit
number-expressions referred to. But those expressions facilitated the transmission of
information from Aaronson to Green, and they did so not because of the efforts that
people made in the way of operating with the rules that assigned them meaning.

More on whether CTM is viable

It is natural to suppose that a language is a set of rules that assign meanings to objects on
the basis of their morphologies. This is exactly what CTM assumes; for it is built into the
CTM-view that there are strictly morphological characterizations of syntactic structures
and also of semantic notions such as proof, entailment, and confirmation.24

24. See Fodor (1987: 19):

“The operations of the machine [on which I am modeling the human mind] consist entirely
of transformations on symbols; in the course of performing these operations, the machine is
sensitive solely to syntactic properties of the symbols; and the operations that the machine per-
forms on the symbols are entirely confined to altering their shapes.”
See also Fodor (1990: 22):
“You connect the causal properties of a symbol with its semantic properties via its syntax…
To a first approximation, we can think of its syntactic structure as an abstract feature of its (geo-
metrical or acoustic) shape [Fodor’s emphasis]. Because, to all intents and purposes, syntax re-
duces to shape [my emphasis], and because the shape of a symbol is a potential determinant of
its causal role, it is fairly easy to see how there could be environments in which the causal role of
a symbol correlates with its syntax. It’s easy, that is to say, to imagine symbol tokens interacting
causally in virtue of [Fodor’s emphasis] their syntactic structures. The syntax of a symbol might
determine the causes and effects of its tokenings in much the way that the geometry of a key
determines which locks it will open.
But, now, we know from modern logic that certain of the semantic relations among sym-
bols can be, as it were, ‘mimicked’ by their syntactic relations; that, when seen from a very great
distance, is what proof-theory is about. So, within certain famous limits, the semantic relation
that holds between two symbols when the proposition expressed by the one is entailed by the
proposition expressed by the other can be mimicked by syntactic relations in virtue of which one
of the symbols is derivable from the other. We can therefore build machines which have, again
within famous limits, the following property:
The operations of the machine consist entirely in transformations of symbols;
 Chapter 13.  Content-externalism and atomism 

We will see in the next chapter that such a view is quite false. Given any language,
actual or possible, an object’s having this or that shape is neither necessary nor suffi-
cient for its being assigned any meaning by that language. But right now, if only for the
sake of argument, let us suppose that this assumption of CTM’s is correct.
What follows from this supposition is not that thought consists in shape-manipu-
lation per se. As we saw a moment ago, sensitivity to semantic (and extra-semantic)
facts is a prerequisite for cognitive achievement, even in domains where all issues can
be decided through decision-procedures. Even where such domains are concerned,
what counts as a genuine cognitive achievement is one’s recognition of a rather com-
plex proposition, one having the form: given that such and such conventions are opera-
tive; and given that, because those conventions are operative, there is a morphological
characterization of truth; and, finally, given that this particular shape has the relevant
morphology; it follows that this particular shape expresses a truth.
There is nothing mechanical about one’s coming to see the truth of such a proposi-
tion, and one’s doing so has no intrinsic connection with one’s manipulating shapes. Of
course, the manipulation of shapes is often a psychological aid to one’s arriving at the
correct judgment; and, further, one often writes down the judgments that one gener-
ates. But in both cases, the connection is entirely causal, as opposed to constitutive,
and is therefore entirely contingent.
Suppose that Smith publishes a proof of some theorem. Here we must distinguish
three things: first, the proof itself; second, Smith’s knowledge of that proof; and third,
his communicating that proof to someone else. The first is an abstract object. The sec-
ond is a psychological condition. The third is a complex state of affairs having both
psychological and non-psychological components. To communicate his knowledge,

in the course of performing these operations, the machine is sensitive solely to syntactic
properties of the symbols;
and the operations that the machine performs on the symbols are entirely confined to alte-
ring their shapes.
Yet the machine is so devised that it will transform one symbol into another if and only if the
propositions expressed by the symbols that are so transformed stand in certain semantic rela-
tions – e.g. the relation that the premises bear to the conclusion in a valid argument. Such ma-
chines – computers, of course, – just are [Fodor’s emphasis] environments in which the syntax
of a symbol determines its causal role in a way that respects its content. This is, I think, a pretty
terrific idea; not least because it works.”
Fodor (1981: 226) writes.
“[C]omputational processes are both symbolic and formal. They are symbolic because they
are defined over representations, and they are formal because they apply to representations in
virtue of (roughly) the syntax of the representations.”
A few paragraphs later, Fodor (1981: 227) writes: “[F]ormal operations apply in terms of
the, as it were, shapes of the objects in their domain.” See also Fodor 1981 (22): “[M]achines
operate on symbols in virtue of their formal (not their semantic) properties…”, and Fodor (1994:
86, 1987: Chapter 1).
 Conceptual Atomism and the Computational Theory of Mind

Smith must produce various noises or ink-marks, and producing such ink-marks and
noises may also help Smith generate that knowledge. (It is easier to solve a problem
with pen and paper than to do so entirely in one’s head.) Symbolic manipulations may
thus be causes or effects of that knowledge. But under no circumstances are they con-
stitutive of it.
CTM seems to involve the view that the behavioral expressions of proof-deriva-
tion are identical with proof-derivation. This view is false, given that it is a form of
behaviorism and that behaviorism is false.25 (Ironically, no one better articulates the
reasons for rejecting behaviorism than Fodor (1968) himself.)26
Some fiction may help both illustrate and corroborate this line of thought. On
some other planet, there is a society S in which symbolic conventions completely dif-
ferent from our own are used. “1” doesn’t mean one, “+” doesn’t refer to addition. In
fact, none of these symbols means anything. In S, a freak accident creates an object X
that is qualitatively just like one of our calculators. Obviously the members of S won’t
regard X as doing anything meaningful. And given how it came into existence, there is
little intuitive support for the idea that X is actually computing anything when it tran-
sitions from “1+4=” to “5.” On the contrary, our intuition suggests that X hasn’t per-
formed a computation or otherwise done anything cognitive. We would have the same
intuition with regard to X’s behavior if it were to have been created, by sheer accident,
on an unpopulated planet. So there is no intuitive evidence that by itself X is computing
or otherwise thinking.
Of course, in our world, when objects like X transition from “1+4=” to “5”, and
make other comparable transitions, many of us have the intuition that those objects
really are computing. But that intuition vanishes when we place X in an environment

25. Chomsky (1959) provides a cogent argument against a doctrine that is called “behaviorism.”
But it is not exactly the kind of behaviorism dealt with by Fodor (1968). Chomsky’s argument is
concerned with behavioristic theories of language-learning, i.e. with views according to which
language-learning is a matter of being behaviorally conditioned to have certain verbal responses
in response to certain sensory inputs. Fodor is concerned with the view that for somebody to have
a mental state of type X is simply for that person to have a disposition to behave in certain ways,
given certain physical inputs. These doctrines are obviously similar, and I believe that some of the
criticisms that Chomsky makes of the one kind of behaviorism can be adapted to apply to the
other kind. But the kind of behaviorism that Chomsky is attacking isn’t necessarily committed to
the view that all mental states are nothing but dispositions towards certain behaviors. An advo-
cate of that view could hold that sense perceptions, for example, are not dispositions of that kind.
26. I would like to propose a psychological hypothesis as to why it is so widely held that thought can
be mechanized. As we discussed in Chapter 1, our intuitions often ascribe too much to our sense-
perceptions. We confuse data with meta-data. We confuse what we see with what we learn on the
basis of what we see. We see a series of innocuous discolorations on a piece of paper or computer-
screen. We happen to know a great deal about the socio-psychological conditions that gave rise to
those discolorations and into which those discolorations are subsequently re-assimilated. Because of
our tendency to confuse perceptual with meta-perceptual data, we end up saying about those disco-
lorations what should really be reserved for those socio-psychological conditions.
 Chapter 13.  Content-externalism and atomism 

where our symbolic conventions don’t hold. This suggests that, so far as people see
objects like X as actually thinking, it is because they are telescoping into their judg-
ments about X’s behavior their background knowledge of the socio-psychological con-
ditions that generated X. In general, so far as people have the intuition that calculators
and other automata think, their intuitions make heavy allowances for the intentions
and social functions of the people and institutions that produce those machines, and
those intuitions are therefore not hewed strictly to the activities of those machines. It
is obvious why these facts strip those intuitions of any credibility that they might oth-
erwise have.
The content-externalist has a counter-response to this last argument
The essence of content-externalism is that what an object is thinking, and there-
fore whether it is thinking, is a function of what its causal liaisons to the environ-
ment are. You are right to say that an object qualitatively just like one of our cal-
culators wouldn’t really be calculating or otherwise cogitating if it were created by
a freak accident on some planet where our symbolic conventions weren’t opera-
tive. But content-externalism shows that an object qualitatively just like you
wouldn’t be thinking about water on a planet where XYZ, as opposed to H2O,
came out of water-faucets and filled our oceans, and so on. Of course, you do have
water-thoughts, the reason being that your mental states have the right causal
origins. By parity of reasoning, a calculator on our planet does have number-
thoughts, the reason being that its operations have the right causal origins.

But we saw in Part I why this point of view cannot be accepted.

Analytic functionalism and the Ramsey-Lewis sentence

There is a way that an advocate of CTM might respond to the general line of thought
just put forth. It involves – indeed, it practically coincides with – a conception of men-
tal content that is widely accepted and that is discrepant with the analysis put forth in
this book. So even though, as we will see, it is not a line of thought that Fodor counte-
nances, it is one that we must examine carefully:
(AF) Given only that machine M outputs a “12” in response to an input of “7+5=”,
it doesn’t follow that M has computed a sum; and you are right to say that one
is guilty of the crudest form of behaviorism so far as one thinks otherwise. But
this is neither here nor there as far as CTM is concerned. David Lewis (1972)
used the concept of a Ramsey-sentence to show how a creature’s mental states
– while not identical with dispositions towards certain behaviors, given cer-
tain inputs – can nonetheless be understood entirely in terms of its behavioral
responses to purely physical inputs. And your argument implodes when
Lewis’ analysis is taken into consideration.
 Conceptual Atomism and the Computational Theory of Mind

Let C be a creature and let S be a giant, conjunctive sentence giving a com-

plete description of C’s psychology. Obviously, S will not contain only psycho-
logical terms; it will contain at least some terms linking psychological states to
publicly verifiable inputs and behavioral outputs. So S will contain sentences
like “ceteris paribus, given a perception of a glass of cold-water, C will grab said
glass and drink its contents.” (Given any psychological theory – whether Freud-
ian, Chomskyan, or Skinnerian – it will surely make at least some allowance for
the behaviors characteristic of mental states, even though those theories may
differ among one another as to the relationship between mental and behavioral
facts. Also, if S doesn’t contain sentences concerning overt behavior, then it is
hard to see how it could constitute a testable, or therefore empirically signifi-
cant, theory. So there is nothing question-begging in Lewis’ supposition that S
will refer to publicly ascertainable inputs and outputs on C’s part.) We now
“Ramsify out” all the expressions in S that refer to mental states. In other words,
given any mental term mi occurring in S, we replace it with a variable xi, and
then prefix the resulting open-sentence with a quantifier that binds all of the
occurrences of xi. (The identity of this quantifier may depend on the value of i.
So it might sometimes be Ffor some xiG, and it might other times be Fgiven any
xi. G) Let S* be the sentence that results from this process.
Relative to any conception of scientific rigor, S* will be no less an adequate
description of C’s psyche than S. Really, the differences between them will be
confined to phraseology. (Where S says “given a perception of ice-water, thirst
is what causes C to reach for the glass”, S* says “there is some state x such that,
given a perception of ice-water, x is what causes C to reach for the glass.”)
Obviously S* contains no mental terms, and the only constants in it refer to
non-psychological inputs and outputs. Unlike behaviorism, S* does justice to
the fact that, at least up to a point, we must understand C’s psychology in
terms of interactions among C’s mental states. But, like behaviorism, S* de-
scribes C’s psychology entirely in terms of stimulus and behavioral response.
And because, as previously noted, any reasonable description of C’s psychol-
ogy will contain at least some sentences linking C’s mental states to its inputs
and outputs, there seems to be no reason why the process just described
wouldn’t hold equally for all admissible values of S.
A moment ago you suggested that CTM involves a simple failure to dis-
tinguish between a mental state and its behavioral expression. Given the con-
cept of a Ramsey-sentence, it seems that this criticism is unjustified. M’s com-
puting a sum isn’t to be identified with its engaging in certain behaviors in
response to a certain input. But its doing that sum is still constitutively linked
with its having a certain kind of behavioral response to that stimulus.
Further, the idea that computing has at least some constitutive connection
with overt behavior seems to me more reasonable, and more scientific, than
your idea that, so far as computing is concerned, behavior is a “mere epiphe-
 Chapter 13.  Content-externalism and atomism 

nomenon.” Your view seems to be that a disembodied brain could compute. But
such a position is inherently untestable, and provides no empirical (observable)
interpretations, no matter how partial, of its theoretical (mental) terms. Given
points that were well established long ago, it follows that your position is incom-
patible with the most basic tenets of scientific methodology.
The position just described is a form of functionalism, and we’ve already (Chapter 12)
seen one reason why that doctrine cannot be accepted.
There are other problems with (AF) and with functionalism generally. One of
these problems is identified by Fodor himself (Fodor 1998: 45, Fodor and Lepore
2002). If content is causal role, then the same content cannot possibly have different
causal roles. But supposing that b is your belief that snow is white, there are obviously
counterfactual scenarios where b has a causal role different from the one it has here. (If
you had written your dissertation on crystallography, instead of on Bentham’s political
theory, your belief that snow is white would have led you to make many inferences that
you have not in fact made.) (AF) makes b’s association with its representational con-
tent much more brittle than it actually is. For similar reasons, (AF) makes it impossible
for two different people to believe that snow is white, since the causal role of one per-
son’s belief in that proposition will inevitably differ from that of another person’s cor-
responding belief. Of course, what we just said about beliefs that snow is white is true
of all beliefs and, indeed, of all propositional attitudes.
The counter-response would be to say that if two causal roles are sufficiently simi-
lar, then they both realize the same content. So causal role fixes content, but content
doesn’t determine a unique causal role.
There is an obvious problem with this counter-response. The causal role of one
person’s belief that snow is white can be extraordinarily different from the causal role
of another person’s belief that snow is white, as the following story shows. Smith is a
physicist who has spent his life studying the reflective properties of crystals and who
believes (correctly, let us suppose) that a solution to the major problems of contempo-
rary physics are to be found through careful consideration of snow’s chromatic prop-
erties. Brown is a ski-instructor who wishes that snow were green, because the glare
caused by snow’s whiteness gives him headaches and makes it hard for him to see on
the slopes. Let bs the brain-state realizing Brown’s belief that snow is white, and let bb
be Smith’s corresponding brain-state.
Obviously the causal roles of these two states differ dramatically. So far as they
have anything in common, it seems to be that each realizes a belief that snow is white:
so far as bb‘s functional role is similar to bs‘s, this similarity lies in the fact that, in each
case, functional role is generated by some object’s being a belief that snow is white. But
if we say that the contents of these states supervene on their functional roles, we can-
not without vicious circularity say that their functional roles are determined by their
 Conceptual Atomism and the Computational Theory of Mind

contents.27 In general, we cannot without vicious circularity understand causal role in

terms of content if we understand content in terms of causal role. (AF) thus turns out
to be viciously circular.28

27. Stich (1983: 79-81) sets forth a similar argument.

28. This brings us to another problem with attempts to understand psychological concepts in
terms of the Ramsey-sentence. Such attempts explicitly liken psychological terms to theoretical
terms. Of course, some psychological terms (e.g. “reaction formation”, “counter-cathexis”, “depth
grammar”) are theoretical. But some (e.g. “tickle”, “feeling of sadness”) are not. The advocate of
(AF) is quite right to say that, for a theoretical term to have meaning, it must have some links,
however circuitous and theory-mediated, with non-theoretical phenomena. And it may (or may
not: see below) follow from this that the meaning of a theoretical term (e.g. “electron”) is at least
partially given by sentences linking the truth-values of sentences about electrons with sentences
describing non-theoretical (observable) states of affairs
But it does not follow that the same is true of psychological terms. Obviously pain, tickles, and
other psychological phenomena have behavioral expressions. But it is decidedly implausible to sup-
pose that the sentences giving these linkages are constitutive of the meanings of terms like “pain”,
“tickle”, and the like. (Wittgenstein disagrees. We will deal with this position in a moment.)
I believe that, so far as philosophers have thought otherwise, it is because they were confu-
sing the term “observable” with the term “publicly observable”, and were then equating the latter
with the term “non-theoretical.” Your pains are not theoretical entities, at least not from your
viewpoint. Further, your pains are observable – they are observable to you, though not to me.
(Of course, you don’t observe them in the sense that you sense-perceive them. But you observe
them in the sense that their existence is made directly known to you, at least in so far as anything
is directly known to anyone.) These facts, by themselves, call into question (AF)’s categorical
assimilation of psychological to theoretical terms.
Most importantly, even though your pains and tickles are known to me only through hypo-
thesis, and are therefore theoretical from my viewpoint, it doesn’t follow that the terms “tickle”,
“pain”, and “sadness” are theoretical terms. Obviously they are not; “pain” is as non-theoretical a
term as we’ll ever find.
So the advocate of (AF) is simply wrong to say that, in order to have empirical significance,
statements containing terms like “tickle” and “pain” must be linked with sentences referring to
publicly observable facts. In any case, he is wrong to say this in so far as his grounds for it lie in
his presumption that terms like “tickle” are comparable to terms like “electron”, on the grounds
that both terms denote “theoretical” and “non-observable” entities.
Also, that claim is false tout court. A brain in a vat could obviously feel pain, joy, and so on.
The advocate of (AF) could counter-respond by saying that such a brain would behave in certain
ways if it were disembodied. Such a counterfactual is true, of course, but only because such be-
haviors would be the effects of those pains and tickles on the host-body. Given that cause and
effect are distinct, that counterfactual presupposes that, in direct contradiction to (AF), pains
and tickles are not to any degree constituted by their behavioral expressions. A similar argument
is found in Pap (1962: 383-393).
Wittgenstein (1958) famously argued that the meaning of words like “pain” and “tickle” is at
least partly behavioral and that such words would be meaningless if their sole function were to
pick out private internal states.
 Chapter 13.  Content-externalism and atomism 

This position involves a failure to distinguish reference-fixing from meaning-giving. “Pain”

refers to pain. It doesn’t refer to behavior, and nothing concerning behavior is any part of its
meaning. Of course, when we wish to tell another person, e.g. somebody learning English as a
second-language, what “pain” means, our doing so often involves our indicating certain beha-
viors. While grimacing or writhing, we say (for example):
(*) “this is pain.”
But the meaning of (*) is given by the statement:
(**) There is some psychological state x such that x causes people to act like this [accompa-
nied by an indication of some stereotypically pain-driven behavior], and “pain” refers to x.
The meaning of (*) is not given by the statement:
(***) “Pain” refers to anything x that causes behavior like this [accompanied by an indi-
cation of grimacing or writhing].
Wittgenstein’s view involves a failure to distinguish between meaning-fixing and reference-gi-
ving and, therewith, to register the ambiguities of scope concealed in statements like (*).
Much of what we just said about beliefs, and mental states generally, is true of all attempts to
“Ramsify out” theoretical terms. Here we must consider an argument given by Hempel (1965:
204-205). So far as its meaning can be given through the procedure that Ramsey and Lewis des-
cribe, the term “electron” ends up being frozen. Every sentence containing the term “electron”
now countenanced by physical science becomes constitutive of the meaning of that term, and is
thus as incapable as any other tautology of being refuted by scientific progress. One might coun-
ter-respond by saying that the Ramsey-sentence is meant only to fix the referent of “electron”, not
to give its meaning. But in that case, the idea that theoretical terms can be “Ramsified out” redu-
ces to the triviality that observation terms help people know what is meant by theoretical terms.
 chapter 14

The concept of a symbol

The purpose of this chapter is to show that CTM involves a spurious understanding of
what symbols are and that, when those misunderstandings are corrected, many of the
classical objections to CTM are vindicated.
If a coffee-machine accidentally produced the sounds “snow is white”, nothing
would have been said; no symbols would have been tokened. Of course, one might
think that the coffee-machine was actually producing symbols. But one would believe
this only in so far as one believed that, in making those sounds, the coffee-machine was
actually animated by certain intentions (or that the coffee-machine had been created by
somebody with certain intentions and that, consequently, somebody was speaking
through the machine, much as one can speak through a letter or an email). The moment
one accepts that the coffee-machine is inanimate and that its production of those sounds
is simply an accident, and not the result of its creator’s intentions, one immediately ac-
cepts that nothing has been said and that nothing linguistic has occurred.1
This suggests that the psychological underpinnings of a symbol-token are not just
causally, but also constitutively, involved in that tokening. Even though it is impercep-
tible to others, one’s mental state is constitutive of the symbol-token one is producing.
When the symbol-type “snow is white” really is instantiated, a constituent of that in-
stance is some kind of mental state. When somebody says “snow is white”, the symbol-
token that has been produced actually comprises at least some aspects of the mind of
the person who produced that token.
Sometimes the psychological aspects of a symbol-tokening are spatiotemporally
remote from its morphological aspects. Suppose that I am typing on a keyboard that
forms letters out of giant rays of light in some remote galaxy. In that case, the morpho-
logical aspects of the token of “snow is white” that I produce are estranged by millions
of years and miles from the psychological aspects of that same token. But we mustn’t
let this lead us astray. The psychological underpinnings of those remote galactic events
are no less constitutive of that sentence-token than those galactic events themselves.

Symbols not strictly psychological entities

At the same time, psychology cannot single-handedly turn an otherwise dead sound into
a living symbol-token. Suppose that, while in the grips of a psychotic delusion, Smith

1. See Searle (1984) for a similar line of thought.

 Conceptual Atomism and the Computational Theory of Mind

utters complete gibberish, thinking that, in English, it means snow is white. In that case,
nothing has been said. Of course, one might be able to figure out that Smith meant snow
is white. But this would be a case of one person’s having psychoanalytic or criminological
insight into another. It would not be a case of two people speaking a common language.
Indeed, it would not be a case of anybody using any language. Let L be any lan-
guage other than English. L may be a public language, or it may be a “private language”
– a code that Smith invented for his own use. Smith’s intention was not to say some-
thing in L – his intention was to say something in English. So even if the noise he made
sounded just like the L-translation of “snow is white”, his making that noise was an ac-
cident, and was thus no more an affirmation that snow is white than the coffee-ma-
chine’s production of that same noise.
Here is a similar argument. Suppose that I try to produce the English words “snow
is white” but that, solely because of a vocal chord malfunction and not because of a
repressed urge to speak French, I end up producing the sounds “la neige est blanche.”
Witnesses would assume that I had spoken a French sentence. But they would be
wrong. It may even be that I do speak French. But given that my French knowledge was
not implicated in my saying “la neige est blanche”, my producing those sounds was not
meaningfully related to anyone’s knowing French, and thus wasn’t a case of French-
speaking. Similar considerations apply to Smith’s case.

Why any symbol-token necessarily has non-morphological components

In order for a symbol to be tokened, it is necessary that it arise in a certain way from a
certain kind of social knowledge. (Later on, we will try to state more precisely the na-
ture of this knowledge.) In this context, the word “necessary” denotes constitutive, not
causal, necessity. So the just-described psychological and social conditions are verita-
ble constituents of the symbol-token.
This doesn’t conflict with our presumption that symbol-tokens, unlike social prac-
tices, are discrete entities. If our analysis is right, a symbol-token is a convergence of a
number of different factors, and that convergence-point is discrete, even though the
converging forces are not.
A symbol-token has morphological, psychological, and sociological components.
Up to a point, but not entirely, the function of the morphological component of a sym-
bol-token is to indicate the existence of the psychological and sociological compo-
nents. Thus, there can no more be a symbol-token in the absence of either of the latter
two components than in the absence of the first. So far as we think otherwise, it is a
reflection of our tendency to telescope metaperceptual into perceptual data. This last
point may require elucidation. We see a mere ink-mark. This is perceptual knowledge.
We know that various social and psychological factors conspired to produce that ink-
mark. This is meta-perceptual knowledge. If it weren’t for this meta-perceptual knowl-
edge, the ink-mark wouldn’t be a symbol-token, at least not for us. So far as we think
 Chapter 14.  The concept of a symbol 

that, in and of itself, that ink-mark constitutes a symbol-token, it is because we believe

ourselves to be perceiving what, in fact, we are meta-perceiving. It is because we are
conflating meta-perceptual with perceptual content.
Here it is crucial to keep in mind the distinction between saying and expressing.
When I produce the sounds “snow is white”, that symbol-token doesn’t mean that such
and such socio-psychological factors are now operative. Semantically, that symbol-
token ascribes a certain color to snow, and that is all it does. But given that a symbol-
token meaning snow is white has been produced, it can be inferred that certain psycho-
social factors are operative. So that symbol-token expresses, but does not state, the
existence of those factors.
Let us sum up. A symbol-token is what results when a bit of morphology is produced
in consequence of a convergence of social and psychological factors. A symbol-token is
thus best thought of as a convergence of psychological and sociological forces. (Later we
will try to give a more precise definition of the terms “symbol” and “symbol-token.”)
Of course, we instinctively gravitate towards the position that symbol-tokens are
discrete physical objects – bits of ink and noise. In fact, that position may even seem a
truism or a definition. But it is only because we labor under a number of misconcep-
tions that we gravitate towards that view (I tried to identify one of those misconcep-
tions a moment ago), and I think that the following comparison may help loosen the
hold that those misconceptions have on us. (The comparison is one that is found in
Wittgenstein (1958). Much of what we will say in connection with money resembles
what is said there.)
What is money? Here is an answer to which we are instinctively drawn:
($) Dollar bills and nickels and quarters are money. They are also physical objects.
They are bits of paper or metal. Money is created (in mints), and it can be
destroyed. In the U.S. it is actually illegal to destroy money. If you burn a dol-
lar bill, you are breaking that law. So a dollar bill is just a piece of paper. For
exactly similar reasons, a nickel is just a piece of metal.
Though there are elements of truth in it, ($) is false on the whole; and it is worthwhile
to say precisely why it is false, since there is such a strict parallelism between ($) and
the conception of symbolhood that I am attempting to refute.
Let B be some particular dollar bill – e.g. the dollar bill that is in your pocket. Now
imagine the following scenario. Tomorrow, the U.S. is taken over by a foreign power. A
new system of government is imposed, and this involves changing the currency that is
used. The bits of paper that Americans previously used to make purchases are now as
useless as Monopoly money.2

2. In his novel The Stand, Stephen King describes a situation where the government collapses
and dollar bills become (as King himself puts it) “as worthless as Monopoly money.” I am bor-
rowing his apt expression.
 Conceptual Atomism and the Computational Theory of Mind

Even if B hasn’t been damaged or otherwise altered, it is now completely worthless

and therefore isn’t money anymore. This shows that B per se never was money. What
once had value, and was therefore money, wasn’t B, but was rather B’s being embedded
in a certain way in a certain set of social practices. B was merely a constituent of some-
thing that had value, and was therefore only a constituent of bona fide money.
Up to a point, B’s purpose was only to indicate the operativeness of the relevant
social conditions.
To be sure, it would be an overstatement to say that B was merely an indicator of the
presence of the just-described social factors. (Let us now assume that the political take-
over earlier described never occurred, and that B remains valid currency.) Suppose that
a copy of a legal contract is destroyed. If a duplicate is found, then the legal situation
hasn’t changed: ceteris paribus the judges, lawyers, and so on, are still going to do what-
ever they were going to do prior to the destruction of the first copy.3 The physical copy
of the contract is there merely to indicate that some kind of agreement has been entered
into; its legal purpose is merely evidential. But if B is destroyed, its possessor has lost
certain entitlements: he cannot get on the bus; he cannot enter the theatre or the ball-
park. It is irrelevant whether B was destroyed in plain view of a judge or public notary.
But in this context what is relevant is that, since money has causal powers that a
mere piece of paper does not, a piece of paper is, at most, a constituent of a unit of
money, the other constituent being some political-legal apparatus. And up to a point,
though not entirely, the function of the piece of paper is merely to indicate the opera-
tiveness of that apparatus.
Analogues of the points just made about money hold of symbols. Noises are partial
constituents of symbol-tokens. If somebody tries to speak, but fails to make a sound, he
has said nothing. But, by itself, a noise isn’t anything linguistic. And when somebody
manages to say something by making a certain noise, the function of that noise is, up to
a point, to show something about the speaker’s attitudes towards certain social rules.
(We will develop this obscure point in Chapter 25.) Hopefully, then, our discussion of
money illuminates, and also corroborates, our contentions about symbols.
Also, our discussion of money brings to light some principles that will enable us to neu-
tralize an obvious criticism that might reasonably be made of our analysis of symbolism:
You may be right to say that, as a matter of analytic necessity, there are genuine
symbol-tokens only where there are socio-psychological factors of the sort you de-
scribe. But you are guilty of a blatant non sequitur in inferring from this that those
factors are veritable constituents of symbol-tokens. As a matter of necessity, I couldn’t
have had different parents from the ones I actually had. But my parents are not in
any sense constituents of me. The idea that a symbol is just a bit of morphology is
thus perfectly compatible with the view (held by you and others) that symbol-to-
kens necessarily originate in socio-psychological conditions of a certain kind. Those

3. To simplify discussion, let us set aside all of the legal niceties relating to the authentication
of duplicates of contracts.
 Chapter 14.  The concept of a symbol 

views are no more incompatible with each other than the obvious fact that people
are distinct from their parents is incompatible with the thesis that no one other than
Y and Z could have been X’s parents if those two are in fact his parents. Your thesis
that psychological and social conditions are actual constituents of symbols thus
seems to be, at best, a proposal that redefine the term “symbol” (or “symbol-token”).
Currently, that term refers to ink-marks and noises; and you are proposing that we
have it refer to ink-marks (or noises) plus their social and psychological concomi-
tants. What you’ve said in connection with symbols is false if taken as an analysis,
and sterile if taken as a proposal to change our nomenclature.

Remember what we said a moment ago about money: we strip money of many of its
causal powers if we identify it with mere bits of paper. Of course, we do not thereby
strip it of all of its causal powers: but we do strip it of those causal powers that distin-
guish it from mere scraps of paper. Like money, symbol-tokens have causal powers. If
we identify symbol-tokens with bits of ink or noise, then we strip them of many of their
causal powers. Of course, we do not thereby strip symbol-tokens of all of their causal
powers: but we do strip them of those causal powers that distinguish them from mere
ink-marks and noises. If we say that a token of “snow is white” is merely a sound or ink-
mark, then we strip that token of the causal powers that distinguish it from the barking
of a dog or the grumbling of someone’s stomach: we strip it of its distinctively symbolic
causal powers. Because so many aspects of human thought and behavior are to be un-
derstood in terms of the manipulation of symbol-tokens, it isn’t reasonable to take the
position that symbol-tokens don’t have any causal powers not also had by mere noises
and ink-marks. Such a position would either nullify obviously correct analyses of hu-
man conduct or it would demand a tedious and unproductive rewording of those anal-
yses. If we are to accommodate the fact that symbol-tokens have causal powers not had
by mere ink-marks and noises, then we must take the position advocated earlier, and
denied by the objector, namely: any symbol-token has a morphological component; but
the function of that component is, at least in part, to indicate the presence of socio-psy-
chological factors that are themselves components of that token.

The concept of morphological elasticity

Consider the following three statements (in this context, the term “ink mark” is short
for “ink mark or noise or pattern of light…”):
(i) An ink-mark is, at most, one component of a symbol-token.
(ii) Whenever an ink-mark is a component of a symbol-token, its function is in
large part to indicate the existence of certain socio-psychological conditions.
 Conceptual Atomism and the Computational Theory of Mind

(iii) Two tokens of a single symbol-type may be morphologically very different

from each other. More precisely, there is no limit to the extent to which two
symbol-tokens may differ from each other in respect of morphology.
We’ve already seen that (i) and (ii) are true. And it is clear that (iii) is a consequence of
(i) and (ii). It is also clear why (i)-(iii) pose a problem for CTM additional to those al-
ready discussed. CTM needs there to be a morphological characterization of certain
semantic notions; it needs to bring semantic facts (e.g. the fact that one sentence entails
another) into alignment with morphological facts. Supposing that our semantic rules
permitted little or no morphological divergences between two tokens of a given sym-
bol-type, such an alignment might be possible (though only to an extremely limited
extent, given various well-known incompleteness-theorems). But given what symbol-
tokens are, the a priori probability of there ever being such an alignment is vanishingly
close to zero, and the conditions under which such alignments would occur would
constitute obfuscations, not expressions, of the basic facts about what symbols are. Giv-
en that any symbol-token has non-morphological components, and that the function of
the morphological component is (up to a point) to indicate the existence of the other
non-morphological components, we cannot reasonably demand of a language that its
morphological aspect remain stable. That would be like expecting the pressure of the
gas in a sealed vessel to remain constant, while we raise its temperature. We cannot
expect dependent variables to behave like constants or like independent variables.
CTM needs the morphological and semantic aspects of language to be in lockstep.
For the reasons just given, the circumstances that permit such alignments are in the
nature of singularities. There are situations where, within limits, a morphological char-
acterization of certain semantic notions are possible. These are cases where what is in
fact the dependent variable (morphology) appears as though it is one of independent
variables (psychology and sociology), thereby creating the misleading appearance that
the socio-psychological components of symbol-tokens are to be understood in terms
of their morphologies. Such situations thus doubly obscure the nature of symbolism,
as they invert the structure characteristic of symbol-tokens, and then proceed to fuse
the ordinarily distinct entities constitutive of that structure. In taking morphology as
primary, CTM erroneously takes such singularities as its paradigms, and erroneously
regards non-singularities as deviations or irregularities.
Some fiction may help clarify my point. In the year, 2025, human beings lose the
ability to speak and to write. They cannot generate new sounds or new ink-marks. For-
tunately, there are still tokens of every letter in the alphabet, and people can form new
utterances out of these. Unfortunately, there is only one token of each letter. All of these
tokens are housed in a universal letter depository. (Since these tokens are all made of
marble, they cannot be bent or warped.) So if I wish to propose to my fiancé, I must go
with her to this letter depository and then physically assemble a set of marble letter-
tokens in such a way that there is a token of the sentence WILL YOU MARRY ME?
 Chapter 14.  The concept of a symbol 

Under these circumstances, the morphological elasticity associated with symbols

of actual languages would be entirely absent. There would be only one way to say KARL
PEARSON DESERVES CREDIT FOR MANY OF THE INSIGHTS ATTRIBUTED
TO THE LOGICAL POSITIVISTS or MARTIN SCORSESE’S MOVIE “CASINO” IS
GROSSLY UNDERRATED.
Under these circumstances, thanks to the complete petrifaction (literally) of symbol-
morphology, it would be possible to give a strictly morphological characterization of at
least some semantic notions. (For reasons given earlier, the “automatization” or “mecha-
nization” of thought would be no more possible under these circumstances than under
our own. But here we are setting this issue aside, so as to develop a new and independent
criticism of CTM.) But notice how much artifice is needed to create the needed align-
ment of morphology and semantics, and notice how unstable that alignment would be.
(It would vanish the moment somebody created new letter-tokens.) This suggests that it
is not in the nature of symbol-tokens to be amenable to such an alignment, and this in
turn confirms our thesis that symbol-tokens are not mere bits of morphology. We’ve al-
ready discussed why this last fact is problematic for CTM.
Let us now develop a point made earlier. It is obvious that, given any physical object
O, and given any meaning (or referent) M, there is some possible language L, such that
O (or, in any case, some object morphologically identical with O) means M in L. But
there is a stronger point to make. Given a language L, and given an object O that tokens
some symbol of L, O’s morphology, taken by itself, completely fails to fix the identity of
the symbol-type tokened by O.
Suppose that Jones, Brown, and Aaron are all competent speakers of English. From
a strictly acoustical standpoint, Smith’s utterance of “we’re having excellent weather
today” may bear much less resemblance to Brown’s utterance of that same sentence
than it does to Aaron’s utterance of “weird squirrels abound in that stack of hay.”
Of course, it would be an overstatement to say that morphology imposes no con-
straints on semantics. Obviously it imposes crucial constraints: if you fail to enunciate,
or if your calligraphy is poor, I won’t know what you are saying. In fact, if your utter-
ances result from an intention to produce the wrong kind of morphology, then your
utterance isn’t a case of speech at all.
Here is an illustration of this last point. Smith produces the sounds “snow is white.”
His intention is to say that grass is green. He says “snow is white” because, not having
a good command of English, he thinks that it means grass is green.
If Smith has affirmed anything, it is either that grass is green or that snow is white.
He hasn’t affirmed that snow is white: he had the wrong intention. He hasn’t affirmed
that grass is green: he made the wrong noise. Therefore, Smith hasn’t affirmed any-
thing. So it is at least partially in virtue of its strictly morphological properties that a
physical event or structure qualifies as a symbol-token.
Nonetheless, as we’ve seen, morphology only has an extremely partial role in the way of
determining whether a physical object betokens this as opposed to that expression-type.
 Conceptual Atomism and the Computational Theory of Mind

Why there cannot be too tight a connection between morphology

and symbolhood

Too much of a connection between semantics and morphology is as threatening to the

existence of a usable language as the absence of such a connection. I have probably never
heard two utterances of “he’s over there” that sound quite the same. The class of circum-
stances that permit a perfect instantiation of a given morphology is only a tiny subset of
the class of circumstances where we wish to communicate the proposition associated
with that morphology, and language would therefore be useless if there were too close a
connection between morphology and meaning.
Consider the deposit of ink to the right of the upcoming colon: beggars can’t be
choosers. It is not often that we can produce a physical object that is morphologically
just like that ink-deposit. But it is often that we wish to affirm the sentiment meant (in
English) by that ink-deposit. Obviously it is easy to express that sentiment: our ability
to do so isn’t diminished by our inability to produce an object having a hyper-specific
kind of morphology. This confirms our point that too much of a connection between
morphology and meaning is as threatening to the existence of a viable language as an
all-out absence of such a connection.
A corollary is that a certain plasticity is built into any aspect of a creature’s func-
tioning that has a genuinely symbolic dimension. To the extent that a creature’s behav-
ior is strictly morphology-based, it doesn’t bear any significant resemblance to any-
thing that is uncontroversially an instance of symbolic activity. Genuinely symbolic
activity does, whereas strictly morphology-driven activity does not, allow for an un-
limited degree of deviation between a creature’s linguistic paradigms and its reproduc-
tion of those paradigms. Genuinely symbolic behavior has a plasticity not had by its
mechanical analogues.
At first it might appear that, if indeed this last point is true, that is merely a reflec-
tion of an evanescent, if not already extinct, fact about our technology. These days one
can buy scanners that have “character recognition.” So if you scan a typed document,
what is uploaded is confined to semantically relevant material, and excludes the se-
mantically irrelevant material (e.g. coffee-stains, crushed insects, smudges) that tends
to gather on documents. A scanner with character recognition can recognize the word
“snow” in physical objects having very different morphologies. So such a scanner
seems to have the plasticity that, according to what was just said, they lack.
Also, there are technologies that transcribe the spoken word. These technologies
are currently imperfect. But the imperfections are rapidly being eradicated; and it is
not unreasonable to expect that, in a few years, there will no longer be a need to type.
Again, it seems that a technological fact counterexamples my analysis.
But this line of thought presupposes the very conception of symbolhood that it is
meant to support and that we have seen reason to reject. Let M be some transcribing-
device that has maximally good character recognition (in other words, suppose that M’s
character-recognition capabilities are as good as they can be, given that M’s behavior is
 Chapter 14.  The concept of a symbol 

strictly morphology-driven), and suppose that M has no trouble “recognizing” the words
that Brown has just uttered, even though Brown has a medical condition that makes it
impossible for him to enunciate properly. Even though it satisfies these conditions, M
will still fail to recognize symbol-tokens that any English-speaker would, in virtue of
being an English-speaker, easily recognize. The following scenario illustrates this.
X and Y are conversing with each other. They are both reasonable people of aver-
age or above average intelligence, and their various sensory modalities are in excellent
working order. At some point in the conversation, Y asks X a question, namely: “what
did Smith do today?” As X and Y both know, and as each knows the other to know,
there are only two answers that, given Smith’s personality and general situation in life,
have any chance of being correct, namely: “he went to O’Reilly’s [the local bar]” and
“he spent the whole day in bed.” Further, X knows that Smith went to O’Reilly’s. Under
the circumstances, it isn’t necessary for X to produce a crisp, lucid utterance of: “he
went to O’Reilly’s.” It is necessary only that X produce a sound that is obviously not a
token of “he spent the whole day in bed.” Such a sound could, and probably would, be
little more than a grunt that has a certain intonation and arc. It could be that X is able
to answer Y’s question with a sound that, in purely acoustical terms, is much closer to
a typical utterance of “we are now having high-tea” than it is of a typical utterance of
“he went to O’Reilly’s.”
But S (the previously described scanner) would wrongly transcribe X’s utterance.
By hypothesis, S’s character-recognition capabilities allow it to respond correctly to
symbol-tokens that are morphologically extremely degenerate. But from a purely mor-
phological standpoint, X’s utterance is so degenerate that even S cannot recognize it.
Even if S can recognize that particular utterance, there will be others that, for ana-
logues of the reasons just given, it will not be able to recognize.
Also, supposing that in that particular case S is sufficiently sensitive to generate the
right symbol-token, that same sensitivity will dispose it to assign the wrong symbol-
token in other cases. There will be cases where S is given precisely the same acoustical
data that it was given here, but where it should generate a token of “we are now having
high-tea” or “Pete likes Green Sleeves.” Because S has the sensitivity needed to generate
the right symbol-token in this case, it will ipso facto be locked into tokening the wrong
symbol (“he went to O’Reilly’s”) in these other cases.
To the extent that context-makes it clear what one is saying, it isn’t necessary that
morphology do so. Obviously morphology always has a role; context can never com-
pletely do the job of morphology. But there is no limit to how much context may mar-
ginalize the role of morphology.
A corollary is that, even if we focus on a single language and even if we leave aside
cases of homonymy, a single morphology can instantiate any number of different ex-
pression-types.
A related corollary is that, for any symbol S and any morphology M that tokens S, we
can imagine a morphology that is different from M that also tokens S. So supposing that
M1…Mn are the various morphologies that, thus far, have tokened the sentence-type
 Conceptual Atomism and the Computational Theory of Mind

“snow is white”, there is a kind of morphology that is not included in that set that could
token “snow is white.” Thus, anything whose operations are strictly morphology-driv-
en is ipso facto incapable of keeping pace with such context-generated variations. For
these reasons, no significant parallel can be established between symbol-driven and
morphology-driven activity.
Of course, there are several objections to be made to this line of thought:
You may be right to say that whole sentences can be tokened with noises that bear
little acoustical resemblance to the paradigms through which English is taught,
and that symbol-morphology deteriorates as context assumes a greater role in
communication. But even after we allow for the morphological deterioration
characteristic of everyday speech, you chose a case of symbol-tokening that is too
degenerate to be regarded as representative and that, consequently, must be re-
garded as a kind of singularity. So far as a hypothesis is counterexampled only by
degenerate or singular cases, it isn’t counterexampled at all. So, given that you can
make your case only in terms of such a tendentious and artificial example, it fol-
lows that you have no case.

A symbol-token has both morphological and socio-psychological components, and

the morphological component has done its job provided that it directs the auditor (or
reader) towards the right social-psychological component. So there is nothing linguis-
tically degenerate about X’s utterance, even though it may fail to satisfy the aesthetic
strictures of a diction-coach.
Of course, X’s utterance is morphologically degenerate, as I myself said. But that is
just an elliptical way of saying that in most contexts an utterance acoustically like X’s
wouldn’t provide the information needed to direct the auditor towards the socio-psy-
chological component of the relevant symbol-token.
There is no sense in which the story just told concerns a singularity or degenerate
case. While it is true that, in the exchange between X and Y, context is doing more than
it ordinarily would, and morphology is doing less than it ordinarily would, these quan-
titative differences don’t constitute structural or categorical differences. We are dealing
with a garden-variety expression of the forces involved in the composition of any sym-
bol-tokening and, therefore, in any linguistic exchange. The dialogue between X and Y
could be described as a “degenerate” or “singular” case only if it were assumed that any
utterance that isn’t in lock-step with the aesthetic strictures of a diction-coach is ipso
facto degenerate. But in addition to being question-begging in this context, such an
assumption would demand that we categorize virtually every utterance ever produced
as degenerate.
Let us consider another objection:
You are confusing statistical with dynamic facts. Technically speaking, there are no
straight lines, rigid bodies, or smooth surfaces. But the physical world must be
understood in terms of such idealizations. So even though they are statistically
absent, they are dynamically or explanatorily present. Statistically there are very few
 Chapter 14.  The concept of a symbol 

utterances of “Smith went to O’Reilly’s” that satisfy the stricture of a diction-coach.

But, in this context, that is irrelevant. What is relevant is that, so far as they dis-
charge successful speech-acts, the bits of morphology that we produce approxi-
mate to such paradigms. X’s utterance of “Smith went to O’Reilly’s” is to be under-
stood in terms of a diction-coach’s crisp utterance of that same sentence. Of course,
what we just said about X’s utterance is true in general. Utterances are understood
only in so far as they are associated with the right paradigms. Such paradigms are
therefore dynamically ubiquitous, even though they are statistically few and far be-
tween. You may be right to say that CTM insists on a much tighter connection
between morphology and semantics than is found in everyday speech. But con-
trary to what you say, that doesn’t show that CTM is wedded to an artificial, or
otherwise erroneous, understanding of what it is for a physical object to be a sym-
bol. On the contrary, it shows that CTM embodies a sensitivity to the relevant dy-
namic facts and, unlike you, doesn’t naively derive semantics from statistics.

I agree with the objector on many points. Just as he says, literal meaning cannot be
understood in statistical terms. (It should be pointed out that Kent Bach 1984 makes
exactly this point.) “I would like a vodka tonic” literally means I would enjoy having a
vodka-tonic right now, even though it almost always communicates the imperative:
bring me a vodka tonic. It seems, indeed, that were complex expressions are concerned,
there is practically no limit to how much literal and understood meaning may pull
apart. (But where simple expressions are concerned, literal meaning must ultimately
collapse into understood meaning.) That is why, as semanticists often point out, there
are well-formed sentences that are systematically misunderstood, for example: “The
man the boy the girl hit kissed moved.”4
But the objector is misapplying this point, as we may see by contrasting the dia-
logue between X and Y with a situation that actually exemplifies the principle that the
objector is discussing. Consider the sentence:
(a) “some number is bigger than every number.”
(a) is false. But it is conceivable that English speakers might take those words to
mean
(b) given any number x, there is some number y such that y is greater than x,
which is true, instead of
(c) there is some number x, such that given any number y, x is bigger than y,
which, being equivalent with (a), is false.
What somebody who wishes to affirm (b) should utter is:
(d) “there is no greatest number.”

4. This example is borrowed from Fodor and Pylyshin (1988: 35).

 Conceptual Atomism and the Computational Theory of Mind

There is indeed a sense in which utterances of “Smith went to O’Reilly’s” should bear a
greater resemblance to a diction-coach’s pronunciation of that sentence than Smith’s
does. But it is not comparable to the sense in which somebody who wishes to affirm (b)
should utter (d) as opposed to (a). If you utter the sentence “Smith went to O’Reilly’s”
with the intention of affirming the proposition Smith went to O’Reilly’s, people are
generally more likely to know that your speech-act is in keeping with the relevant lin-
guistic rules if your utterance is more like a diction-coach’s than it is like Smith’s. It is
easier to know what a diction-coach is saying than it is to know what is being said by
somebody who is mumbling or has laryngitis. But this doesn’t mean that the expres-
sions produced by a diction-coach are more correct, relative to the rules constitutive of
the language in question, than what is said by somebody who has laryngitis. It only
means that ceteris paribus it is easier in the one case than the other to identify the
speech-acts that have in fact been performed. So the sense in which one should speak
like a diction-coach is comparable to the sense which one should speak directly into a
microphone. The “should” here is of a generally instrumental nature, and has no spe-
cifically linguistic significance.
But supposing that I wish to affirm (b), the sense in which I should utter (d), as
opposed to (a), is of a distinctively linguistic nature. In uttering (a), I am guilty of lin-
guistic incompetence. But I am guilty only of garden-variety instrumental incompe-
tence if I mumble (d) with the intention of affirming (b).
There is another objection to our analysis that we should consider:
You start with a correct point and end with a false one. There are indeed many
ways that a given symbol-type of a public language can be instantiated, and natu-
ral languages are thus characterized by a certain “morphological elasticity.” As you
point out, this makes it harder than it would otherwise be to bring morphology
into alignment with semantics. But this is completely irrelevant to CTM. Accord-
ing to that doctrine, thinking consists in manipulating symbols of some internal
language, not of some public and external language like English or Spanish. To
immunize CTM from the criticisms that you’ve just put forth, we need only as-
sume that this internal language (“Mentalese”) is petrified, i.e. that where that
language is concerned, there are strict limits as to how much two instances of a
given symbol-type can differ from each other in respect of morphology.

This line of thought is a non-starter, the reason being that it is inherent in the nature of sym-
bolism that morphologically different entities be capable of betokening the same symbol,
and also that morphologically identical entities be capable of tokening different symbols.
Whether a given sound (or ink-mark or pattern of light….) betokens this as op-
posed to that symbol is a function, not of that sound’s morphology, but of the relation
between its morphology and the socio-psychological context of tokening. That is why
two morphologically identical sounds (or ink-marks or patterns of light…) can beto-
ken different words, and that is why two morphologically very different sounds (or
ink-marks…) can betoken the same word.
 Chapter 14.  The concept of a symbol 

Let T1 be an occurrence of the word “bank” (financial institution), and let T2 be a

morphologically identical occurrence of the word “bank” (river’s edge). Why do T1
and T2 betoken different words? In the one case, the context of utterance involved
Smith putting a fishing pole into the back of his pick-up truck. In the other case, the
context of utterance involved Jones angrily snapping his briefcase shut after unsuccess-
fully attempting to borrow money from his brother. Even though the morphology is
the same, the relationship between morphology and context of utterance has changed;
and that is why two words have been tokened, instead of one. Obvious extensions of
this reasoning show why two different tokens of a single expression-type may be very
different in respect of morphology.
This line of thought accommodates the fact that morphology has some constitu-
tive involvement in the tokening of a symbol. For a given symbol to be tokened, it is
necessary that there exist a physical object (e.g. a noise or ink-mark) whose morphol-
ogy stands in a certain relation to the context. This means that there is no symbol (or
symbol-token) where there isn’t a bit of morphology. It also means that, all other things
being equal, two tokens of a given expression-type will be morphologically similar.
But this same line of thought accommodates the fact that two tokens of a given
expression-type can be morphologically very different. Given a sufficiently large
change in the context of utterance, a commensurately large change in morphology will
be needed to ensure that the relation between morphology and context is preserved. (If
you gain two hundred pounds, the clothes that you wear must also change if they are
to fit, i.e. if their relationship to your new physique is to coincide with the relationship
had by your old clothes to your old physique.)
A symbol-type is instantiated in a given context iff there occurs an event whose
morphology has a certain relation to that context. A symbol-token is thus an instance
of a certain kind of relation: a relation between morphology and context. A symbol-
type is such a relation. For reasons already discussed, the word “context”, as it occurs
here, is to be understood, at least up to a point, in social and psychological terms.
Because symbols are relations between morphology and context, it is not an option to
suppose that there are, or even could be, morphologically petrified languages. Here is a
stark illustration of the principle in question. A ratio is, by definition, a relation between
two integers. So the ratio 1:2 is identical with the ratio 3:6, since the relation between one
and two is the same as the relation between three and six. Suppose that somebody de-
cided to speak of “numerator-frozen” ratios. Given two integers m and n, m:n would be a
numerator-frozen ratio if there were no integers m* and n*, distinct from m and n, such
that m:n was identical with m*:n*. The concept of a numerator-frozen ratio is obviously
an absurd one. Given that symbols are relations between morphology and context, one’s
thinking involves a similar absurdity so far as it presupposes the existence, or even the
possibility, of morphologically frozen languages.
We’ve seen some reason to think that the concept of a symbol-token is psycho-
logically pregnant. Much of the debate between advocates and opponents of CTM con-
cerns whether this is actually true and, if so, whether it has any bearing on the thesis
 Conceptual Atomism and the Computational Theory of Mind

that thinking consists in symbolic-manipulations. But the criticism that we just put
forth is independent of this debate, and doesn’t presuppose the truth of any particular
resolution of it. Given that a symbol-token is an instance of a relation between mor-
phology and context, it immediately follows that there is no purely morphological
characterization of any symbol-type. It immediately follows that there is no strictly
morphological characterization of any of the symbol-types constituting any language,
whether actual or possible. Of course, we’ve seen reason to believe that the contexts in
question have a psychological (and also a sociological) dimension, and advocates of
CTM are likely to deny this. But the argument just presented has nothing to do with
the specific nature of the contexts in question, and it goes through regardless of wheth-
er we were right to see such contexts as being psychologically pregnant. Once it is
granted that a symbol-token is an instance of a relation between morphology and con-
text, then regardless of whether that context is understood in psychological terms it im-
mediately follows that there is no one morphology that two tokens of any symbol-type
of that language must have in common. More exactly, given any symbol-type S of any
language L, if two physical events are morphologically similar, it cannot be solely in
virtue of their both being instances of S. So supposing that DOG is a symbol-type of
Mentalese, if two brain-events are morphologically similar, that is not solely in virtue
of their both being instances of that symbol-type; and given only that two brain-events
are both tokens of that symbol-type, there is no limit to how much they can differ in
respect of morphology.
Thus, symbolic manipulations never track semantics in virtue of being morpholo-
gy-driven. This must be understood aright. Obviously morphology-driven operations
sometimes track semantics. A monkey can manipulate letter-blocks in such a way that
they form a meaningful sentence-token. But given any language, and given any series
of inscriptions (or noises…), the morphologies of those inscriptions leave it entirely
open what, if any, expressions of that language have been tokened. Because symbols
are relations between morphology and context, we cannot even coherently conceive of
a language of which that isn’t true.
It could well be that, circumstances being what they are, there is in fact only one
way to token the symbol-types of a particular language. We described such a scenario
earlier. But it can never be inherent in any language that its symbols are thus frozen.
Conceivably, an advocate of CTM might respond by saying that, even though
Mentalese expression-types can in principle be tokened by different morphologies, the
structure of the brain prohibits different morphologies from instantiating a given sym-
bol-type – much as, in our story about the letter-depository, contingent circumstances
prohibited different morphologies from instantiating any given English word.
But, in light of what we’ve said, it is clear why this move won’t work. Because it is
never strictly in virtue of its morphology that a physical object is a token of some ex-
pression-type, it is no longer an option to say that it is ever strictly in virtue of the fact
that shapes are being manipulated that a symbolic operation, or therefore a computa-
tion, is being performed. Suppose that, because he just ate a rotten fish, Harry becomes
 Chapter 14.  The concept of a symbol 

violently ill. In that case, the cause of Harry’s becoming ill coincides with his eating a
certain fish. But it is in virtue of his eating something rotten, and not in virtue of his
eating a fish, that Harry becomes ill. To put it another way, if X is the fish in question, it
is not X’s being a fish eaten by Harry, but rather X’s being a rotten thing eaten by Harry,
that leads to Harry’s illness. (We must remember that it is states of affairs, not objects,
that have causal powers.) Similarly, even though interactions among bits of morpholo-
gy may coincide with interactions among symbol-tokens, nothing that is symbolic oc-
curs in virtue of anything that is strictly morphology-driven: because it is never solely
in virtue of its morphology that anything is a symbol-token, no strictly morphology-
driven interaction of symbol-tokens is itself symbolic. Thus, given two symbol-tokens
A and B, unless it is (at least in part) A’s having certain non-morphological properties
that brings about B, the juxtaposition of those two symbol-tokens does not itself consti-
tute anything symbolic. So, supposing that our thoughts are mediated by symbol-to-
kens, it must be the non-morphological properties of such tokens that determine in
what sequences they occur. Otherwise those sequences, though consisting of symbol-
tokens, will not themselves be symbolic. And in that case, those sequences will consist
of thoughts that do not add up to larger thoughts, the result being a complete absence
of thought-processes of any kind, whether computational or not.

Why CTM strips the mental of any causally significant properties

A consequence of our analysis is that, in direct opposition to what Fodor believes, CTM
is inconsistent with the fact that mental states have causal powers. In fact, CTM makes
it harder than it would otherwise be to accommodate the obvious fact that it is often in
virtue of their representational properties that mental states have causal efficacy. I
would now like to say, as explicitly as possible, why this is so. (Doing this will involve
repeating some points made earlier.)
Here is Fodor’s view:
To think is to compute. To compute is to manipulate symbol-tokens. Symbol-to-
kens are shapes. Shape is causally efficacious. (In virtue of being round, as op-
posed to square, an object has causal properties that it wouldn’t otherwise have.
That is why square wheels would be so useless.) So if we identify thinking with
computing, we have no trouble accommodating the fact that our thoughts have
causal powers – that thoughts generate other thoughts and also generate actions.

An expression-type is not a shape, and an expression-token is not an instance of a

shape. An expression-type is a relation between morphology and context, and an ex-
pression-token is an instance of such a relation. So to identify thinking with comput-
ing is to identify thinking with the manipulation of instances of relations between
morphology and contexts. (More exactly, it is to identify thinking with the manipula-
 Conceptual Atomism and the Computational Theory of Mind

tion of instances of relations of instances of morphology to instances of certain kinds

of context.) It is not to identify thinking with the manipulation of bits of morphology.
Suppose that S is a token of the Mentalese word SOCRATES. If our analysis is correct,
S is an instance of a relation between morphology and context. Given CTM, the morphol-
ogy in question must be had by some brain-state B. But what is the relevant context?
There are different possible answers to the question, and none of them validates
CTM. Let us start with the answer that Fodor himself gives. According to Fodor, so far
as B represents Socrates, it is because some state of affairs of which Socrates was a part
is causally responsible for B’s existence. So if Fodor is right, the relevant context is
some long-gone state of affairs of which Socrates was a part. In that case, any manipu-
lation of S would involve a manipulation of some vast stretch of human history. But it
obviously makes no sense to say that vast stretches of history are manipulated. In any
case, thinking obviously doesn’t consist of such manipulations. It should also be point-
ed out that in holding that thinking consists in manipulations of symbol-tokens, CTM
is committed to the most aggressive form of content-externalism imaginable. This is
because the symbol-tokens in question would involve remote stretches of space-time,
and CTM therefore becomes no more capable than any other form of content-exter-
nalism of explaining how our thoughts can have any causal properties at all.
Fodor’s view is that B – the intra-cranial bit of neuro-morphology – is the relevant
symbol-token and that, consequently, there is no trouble explaining how it is that B
can be manipulated. But, given what the symbols are, CTM cannot coherently say that
B itself is the relevant symbol-token.
We’ve seen that Fodor’s answer to the question “what is the relevant context?” doesn’t
enable CTM to accommodate the fact that thoughts are not causally inert (i.e. that, in
virtue of being a thought, a brain-event has causal properties that it wouldn’t otherwise
have) and, further, that in virtue of being a thought with a specific content C, a brain-
event has causal properties that it wouldn’t have if it instead had some other content. But
is there some other possible answer that would accommodate this fact?
Obviously such a symbol-token would have to be entirely intra-cranial. Otherwise,
as we just saw, we are forced to say, absurdly, that thinking consists in manipulating vast
stretches of extra-cranial space-time. It immediately follows that any version of CTM
that can accommodate the fact that thoughts have any causal powers is completely in-
compatible with content-externalism. This is a problem for Fodor, since one of his prin-
ciple reasons for advocating CTM is that he accepts content-externalism and believes
that an acceptance of CTM is needed to reconcile content-externalism with the fact that
thoughts are causally efficacious. But does it follow that CTM itself is in jeopardy?
Yes. As we’ve just seen, one must accept content-internalism if those symbols (or
symbol-tokens) are intra-cranial. One must take the view that, if a brain realizes men-
tal states with certain contents, that is to no degree a function of the causal origins of
that brain’s current condition, and is thus entirely a function of those properties of that
brain that can be understood in abstraction of those origins. But if one takes that view,
then there is no need to deny that it is in virtue of their representational properties that
 Chapter 14.  The concept of a symbol 

brain-states have causal powers, and (if like Fodor, you are an advocate of SCT) there
is thus no need to deny that it is in virtue of their semantic properties that brain-states
have causal powers. We can continue to say what we have a strong pre-theoretical wish
to say, namely: beliefs and desires have causal powers (or, more precisely, that if x is a
thought with a certain content, then in virtue of that fact x has causal powers that it
wouldn’t otherwise have).
In identifying cognitive operations with non-semantic operations, CTM demands
an abandonment of the our presumption that my desire to drink water is at least part
of what causes me to walk over to the water fountain and drink from it. In fact, an
abandonment of that presumption is really the essence of CTM, since that doctrine is
given by the thesis that semantics does nothing and morphology does everything.
We’ve just seen that if symbol-tokens are intra-cranial, there is no need to deny this
presumption, and thus no need to accept CTM. We’ve also seen that CTM fails if Men-
talese symbols are, even in part, extra-cranial. So there isn’t any delineation of the term
“symbol” that validates CTM.
Let us consider a point that might be made on behalf of CTM:
Expressions of English are tokened by bits of noise or ink, whereas expressions of
Mentalese are tokened by patterns of neural stimulation. Tokens of English expres-
sions are typically audible (or visible), whereas tokens of Mentalese expressions are
not. One has to learn English, but one doesn’t have to learn Mentalese. These are but
a few of the many differences that we may expect to find between Mentalese and any
public language. No advocate of CTM has ever denied that such differences abound,
and none of these differences warrants the rejection of CTM. Consider the fact that,
if they exist, tokens of Mentalese expressions aren’t made of ink or noise. It would be
absurd to conclude on this basis alone that thinking does not consist in operations
on symbol-tokens belonging to an internal language. It seems to me that all of your
criticisms of CTM are comparably absurd. All you’ve done is to uncover more evi-
dence in favor of the trivial point that Mentalese differs from English (and other
public languages) in many ways. Leaving aside your view that CTM is wrong, you
haven’t said anything that any advocate of that doctrine would necessarily disagree
with. At most, you’ve helped CTM hone its conception of the nature of the internal
symbols that mediate thought, by eliminating wrong views about those symbols.

Consider the English word “dog.” Here is what we have said about it. If a noise tokens
that word, it is not solely in virtue of its morphology. Rather, it is in virtue of its mor-
phology plus facts about the context of utterance. More exactly, it is in virtue of the
relationship between the morphology of that noise and the context in which it oc-
curred. There are instances of expression-types when, and only when, there are in-
stances of relations of the sort just described. This suggests (as we previously observed)
that, where public languages are concerned, expression-types just are such relations
and expression-tokens are instances of such relations.
 Conceptual Atomism and the Computational Theory of Mind

Supposing that this analysis is correct, anything that isn’t an instance of such a re-
lation doesn’t have any significant similarity to the things that are ordinarily referred to
as “expressions.” We’ve discussed why CTM cannot coherently regard tokens of Men-
talese expressions as instances of relations of this kind. CTM thus cannot coherently
regard cognition as an operation on anything that can appropriately be referred to as
an “expression” or “symbol.”
We must make a distinction. It is one thing to say that there are no interpretations
of terms like “symbol” and “linguistic expression” that validate CTM, and it is quite
another to say that we don’t think in symbols or linguistic expressions. What we’ve
seen so far is not that we don’t think in symbols. What we’ve seen is that there are no
interpretations of expressions like “symbol” and “sentence-token” that validate CTM’s
thesis that thinking consists in formal symbolic operations. We have not yet seen that
thinking doesn’t consist in the tokening of expressions of some kind or other.
CTM and SCT are distinct doctrines. CTM presupposes the truth of SCT, but SCT
doesn’t presuppose the truth of CTM. I will argue that SCT is indeed false. But it is not
false because CTM is false; it is false for its own distinctive reasons.

Mental causality revisited: Fodor’s syntactical approach

For reasons that we discussed in Part I, content-externalism seems to be incompatible

with the causal potency of the mental. We discussed a number of attempts to refute
this, but found them to be fallacious. But there is one important approach to this prob-
lem that we did not discuss – that of Jerry Fodor.
Fodor’s analysis makes heavy use of the concept of syntax. We were not in a posi-
tion to evaluate Fodor’s analysis in Part I, since we hadn’t yet analyzed that concept.
Now that we have done so, we can assess the merits of Fodor’s view.
As we’ve discussed, Fodor is a hard-line content-externalist. In his view, given only
that, leaving aside facts about the causal origins of their conditions, Max and Twin-
Max are qualitatively identical, it doesn’t follow that they have anything in common in
terms of the representational contents of their mental states. At the same time, Fodor
rightly takes it for granted that psychologically they are indistinguishable, i.e. that what
an omniscient psychoanalyst would (in his capacity as psychoanalyst) have to say
about the one would be identical with what he had to say about the other. Fodor’s posi-
tion is thus consistent with the causal efficacy of the mental and, therefore, with the
presumption that psychology has integrity as a discipline.
Fodor reconciles these two views by saying that, whatever causal properties a
thought has, it has them in virtue of its syntax, not its semantics (representational con-
tent). According to Fodor, given that Max and Twin-Max are atom-for-atom duplicates
of each other, the syntactic structures of Max’s thoughts must be identical with those
of Twin-Max’s thoughts. What those thoughts represent is a function of spatiotempo-
rally remote, and therefore causally inert, facts about the environments in which those
 Chapter 14.  The concept of a symbol 

individuals are embedded. So the semantics of a person’s brain-states are to be under-

stood in terms of what is remote and therefore causally inert. But syntax is an entirely
internal affair. Given two subjects that are atom-for-atom duplicates, their thoughts
cannot differ in respect of syntax, it being irrelevant what environmental facts led to
those thoughts.
Is this position tenable? No. First of all, thoughts have syntactic structure only if
they are sentences. Obviously this is not an innocuous view (later we will find it to be
false). But for argument’s sake, let us suppose that it is correct and that mental states do
indeed have syntactic structure. Given this, let P be the meaning of
(a) “Mary loves Tom”,
and let P* be the meaning of
(b) “Larry punched Bob.”
Further, let
(A) Mary loves Tom,
and
(B) Larry punched Bob
be tokens of the Mentalese translations of (a) and (b).
There is no denying that (A) and (B), being brain-states, have causal properties.
They have mass, shape, temperature, and various other causally efficacious properties.
The question is not whether brain-states have causal powers or whether Fodor’s view
strips them of such powers. The question is whether, supposing that Fodor is right to
strip semantics of causal power, he can coherently say that syntax has causal power. In
other words, can syntax be causally efficacious if semantics is not?
No. As we saw earlier, a sentence-token’s syntax lies in how it means what it means.
Syntax, as we saw, is meaning-how (whereas semantics is meaning-what), and mean-
ing-how no more supervenes on morphology than does meaning-what. Given only its
morphology, a token of (a) doesn’t have to mean Mary loves Tom, and the same is obvi-
ously true of (A).
An advocate of CTM could respond by saying that, where Mentalese symbol-to-
kens are concerned, syntax does supervene on morphology. But given what we said in
the last chapter, this would mean that Mentalese symbol-tokens (so-called) were so
different from their English and Spanish counterparts that we couldn’t refer to the
former as “symbols” without rendering that term ambiguous.
 Conceptual Atomism and the Computational Theory of Mind

The Regress-argument revived

We saw in Chapter 13 that syntax is semantic decomposition. A sentence’s syntactic

structure lies not in what it means, but in how it is assigned that meaning by the se-
mantic rules of the language to which it belongs.
Supposing that our thoughts have syntactic structure, the relevant semantic rules
are either mentally represented or they are not. If they are mentally represented, those
representations cannot themselves coincide with Mentalese sentences (or sentence-
tokens), since (for well known reasons5) no language can “contain its own truth-predi-
cate”, as Tarski put it. (In other words, no consistent language can express all of its own
semantic rules.) Those rules must therefore be represented either in some non-linguis-
tic form or in Meta-Mentalese (a language distinct from Mentalese that expresses the
rules that assign meaning to Mentalese expressions). If we say that those rules are rep-
resented in a non-linguistic form, then we are giving up on the thesis that we think in
sentences. If we say that they are represented in sentences (or sentence-tokens) of
Meta-Mentalese, then we embark in a vicious regress, since everything that we said
about Mentalese is true of Meta-Mentalese.
So let us suppose that the semantic rules of Mentalese are not mentally or neu-
rally or otherwise internally represented.6 In that case, the operativeness of those rules
does not supervene on any causally effective fact about one’s person. But then the syn-
tactic structures of Mentalese sentence-tokens are stripped of any causal powers. It
straightforwardly follows that Fodor is wrong to hold that it is ever in virtue of their
syntactic properties that brain-states do any causal work.
In light of these points, we can see why one of the classic arguments against SCT
is in fact cogent. If we think in a language, then presumably we understand that lan-
guage – we know how to interpret its sentences. But interpreting such a sentence either
involves translating it into another language that is already understood or it involves
generating a non-linguistic representation of its meaning. In the first case, there is a
vicious regress. In the second case, there is an acknowledgement that thinking ulti-
mately does not consist in the tokening of sentences. Either way, SCT fails.
This argument is well-known to advocates of SCT, and here is how they deal with it:
(NSK) To know a public language, like English, you must mentally represent its
rules. It would indeed be viciously regressive to suppose that one must mentally
represent the rules of Mentalese. But Mentalese is not a public language; and to
circumvent the argument that you just put forth, advocates of the view that we
think in Mentalese symbols need only say that you don’t have to know the seman-
tic rules of Mentalese to know Mentalese. All that is necessary is that, as Lycan

5. See Kuczynski (2002, 2005) for a detailed discussion of these reasons, and Russell (1950:
371) for a pithy and lucid discussion of them.
6. This is the position taken by Fodor (1975) and Lycan (1984: 237) in response to the regress-
argument just given.
 Chapter 14.  The concept of a symbol 

(1984: 247) puts it, you be “built to conform” to those rules. You use English-ex-
pressions in the right way because you know how to use them. But even though
you use Mentalese expressions in the right way, it is not because you know, or
otherwise mentally represent, the corresponding semantic rules. It is because your
nervous-system is so structured that you are physically compelled to token those
expressions in the appropriate ways. The compulsion here doesn’t involve the me-
diation of semantic, and therefore regress-generating, activity.7

To quote Fodor himself:

[Y]ou don’t have to represent the rules of a language that you are able to use: all
that’s required is that you are so constructed as to conform to the rules. (Cf. the
way computers are ‘built to use’ the machine language.) Of course, it’s plausible
that you DO internally represent the rules of (say) English; but that’s not because
English is a language; it’s because it’s a language that you LEARN; and it’s quite
plausible that to learn a language is to come to internally represent its syntax and
semantics (presumably in ‘Mentalese’).8

If (NSK) is right, Mentalese is so different from any paradigm-case of a language that

we must once again question whether it is appropriately described as a language. The
answer to that question is “no”, as we will now see.
Given any one English-speaker, the English-language could exist if that person didn’t
exist or did exist but didn’t speak English. But suppose that nobody spoke English. In that
case, the English language wouldn’t exist. One might respond by saying that languages
are purely function-theoretic pairings of physical objects with meanings. In a moment,
we will consider this position, finding it to be false. But even if it is true, it remains an
uncontroversial fact that, if nobody spoke English, English would be defunct and thus as
good as non-existent.
Knowing English consists in knowing the relevant semantic rules. English is dead
if nobody knows those rules. It is obvious that, in general, whether a public language
is alive or dead is constitutively dependent on whether people know the relevant se-
mantic rules. (This doesn’t mean that any one person has to know all of those rules, but
somebody must know some of them.) It is an essential fact about any public language
that its very existence (or, in any case, its not being dead) is constitutively dependent
on its rules’ being known to somebody.
But for a language to be alive, it is not sufficient (though it is necessary) that some-
body know its semantic rules. A knowledge of those rules must be causally responsible
for the judgments that people make as to what noises and ink-marks mean, as the fol-
lowing story shows. You privately learn how to read and write some long-dead lan-
guage L, and a number of other people simultaneously do the same. But nobody knows
that anyone else knows L, and nobody uses their knowledge of L to communicate any-

7. See Lycan (1984: 247), Kuczynski (2002, 2004d).

8. Fodor, private correspondence. The capitalizations are Fodor’s own.
 Conceptual Atomism and the Computational Theory of Mind

thing to anyone else, or even to themselves. Under these circumstances, L is obviously

a dead language (though it could be readily revived). So supposing that E1…En is a
complete list of the expression-tokens that people produce, a language L is alive only if,
for some i, a knowledge of L it is causally responsible for somebody’s producing Ei.
(Obviously this is not a sufficient, but only a necessary, condition for L’s being alive.)
If (NSK) is right, then nobody knows the semantic rules for Mentalese. A fortiori
a knowledge of those rules is inert. (Even if Mentalese does exist, and a few cognitive
scientists manage to figure out what those rules are, that knowledge obviously isn’t to
any degree what sustains the continued existence of Mentalese. The fact that a few
people know Sanskrit isn’t enough to make Sanskrit be a living language.) Mentalese is
therefore crucially different from anything that is uncontroversially a language.
A related point is that, as a matter of analytic necessity, where a knowledge of seman-
tic rules is inert, there are no semantic rules, at least no living ones. If (NSK) is right, the
semantic rules of Mentalese exist, and remain active, even though nobody knows any of
them. For this reason, the so-called semantic rules associated with Mentalese are so dif-
ferent from those associated with English that it is misleading to use the term “semantic
rule”, without qualification, in connection with both representational systems.
How are the (so-called) semantic rules of Mentalese relevantly similar to those of
English? Like their English counterparts, the (so-called) semantic rules of Mentalese as-
sign representational content to bits of matter (or events). But if anything that assigns
content to a bit of matter is ipso facto a semantic rule, it becomes trivial to say that our
thoughts are linguistic, and the thesis that we think in symbols is reduced to the platitude
that we have thoughts that have content or, more simply, that we have thoughts.
But even this overstates the similarity between Mentalese semantic rules and their
English counterparts. English semantic rules have normative content. They are to the ef-
fect that, if one wishes to affirm such and such, one should utter thus and such (or, at least,
that one can do so). It is a semantic rule that “pickles like to listen to harpsichord-musirc”
means pickles like to listen to harpsichord-music. But that rule is not to the effect that
anyone in fact uses that sentence to convey that proposition. That rule is to the effect, if
one wishes to convey that proposition, one ought to use that sentence (or, at least, that one
can do so). But given the Fodorean assumption that Mentalese semantic rules are un-
known, those rules are without normative content: they are not to the effect that, if one
wants to convey such and such, one ought (or can) do so by doing thus and such. Those
rules are to the effect that such and such bits of matter in fact have thus and such content:
unlike English semantic rules, they are prescriptively and permissively empty. This con-
firms our point that there are no meaningful similarities between the things that we know
to be semantic rules, on the one hand, and the semantic rules of Mentalese, on the other.
(NSK) thus amounts to nothing less than a rejection of the SCT. Supposing that
(NSK) is right, the only significant property that Mentalese expressions have in com-
mon with their English (or Korean or French…) counterparts is that, in both cases, the
expressions (so-called) in question have content. Any other putative similarity ends up
being skewered on an extension of the line of thought just put forth.
 Chapter 14.  The concept of a symbol 

For example, suppose that one were to suggest, as Fodor does, that Mentalese ex-
pressions, like their English counterparts, have syntactic structure. So far as English ex-
pressions have syntactic structure,that it is because somebody knows at least some of the
rules of English semantics. We’ve already seen that a knowledge of syntax is indistin-
guishable from a knowledge of the kind of semantic information that is given by contex-
tual definitions. One knows English syntax because one knows (inter alia) how “loves”
combines with other expressions into meaningful units – because, in other words, one
knows the relevant recursive semantic rules. Syntactic knowledge is combinatorial se-
mantic knowledge. We’ve seen that there would be no English semantic rules if nobody
knew any such rules. The same is therefore true of English syntactic rules, given that
syntactic rules are (combinatorial) semantic rules. These points mutatis mutandis hold
with respect to the syntactic rules of anything that is clearly a language. So if we say that
there could be Mentalese syntactic rules even if nobody knew any of them, we are saying
that Mentalese is so different from anything that clearly deserves to be described as a
language that it does not itself deserve to be so described.
There is another problem with Fodor’s syntactical approach. Let us look at our
paradigms again (as before, suppose that (A) and (B) are the Mentalese translations of
(a) and (b)):
(a) “Mary loves Tom”,
(b) “Larry punched Bob.”
(A) Mary loves Tom,
(B) Larry punched Bob
Since (a) and (b) are syntactically identical, the same is presumably true of (A) and (B).
Supposing that Fodor is right to say that there are mental-causal differences only where
there are syntactic differences, it follows that ceteris paribus one’s belief that Mary loves
Tom has the same causal properties as one’s belief that Larry punched Bob. Since this
consequence is false, so is Fodor’s analysis.
Because he identifies syntax with morphology, and not with semantic decomposi-
tion, Fodor would not be moved by these arguments.9 But syntax is not morphology,
as we have seen, and the argument just given therefore stands.
One might object that, for special reasons, the Mentalese translations of (a) and
(b) don’t have the same syntactic structures. But what we just said about (A) and (B)
will hold of any two Mentalese sentence-tokens that are semantically different but syn-
tactically identical.
In response, an advocate of SCT might say that, where Mentalese is concerned, it
is impossible for two sentences to have the same syntax but different semantics. (This
is not far from the position that Fodor himself sometimes appears to take, since he
sometimes says that there are syntactic differences whenever there are morphological

9. See, for example, Fodor (1987: 18-20). See also Cain (2002: 135).
 Conceptual Atomism and the Computational Theory of Mind

differences.) But this suggestion is a non-starter. Syntax is recursive semantics. If Men-

talese satisfied the condition just described, the same recursions could never be used
twice. Since anything comparable to English or Korean – anything that clearly deserves
to be described as a “language” – just is a set of recursions defined over a certain lexi-
con, that is tantamount to saying that there is no such language as Mentalese.
I would like to end this chapter with a general point about symbolism. We saw
earlier that symbol-types are relations between morphological and psychosocial fac-
tors, and that symbol-tokens are instances of such relations. (By a “psychosocial fac-
tor” I mean a case of somebody’s having a belief about, or at least some kind of propo-
sitional attitude towards, some social institution, e.g. a convention of some kind.) But
not just any such relation qualifies as a symbol, for there are obviously non-symbolic
relations holding between morphological and psychosocial factors. Right now, while
sitting at my desk, I am contemplating the fact that there is little crime in Sweden.
Surely this confluence of circumstances isn’t an instance of some symbol’s being to-
kened, even though it is an instance of a relation between morphology and psychoso-
ciology. So in saying that symbols are relations of the kind just described we are giving,
at most, a partial analysis of the concept of symbolism. What is the right complete
analysis of that concept? We will outline an answer to this question in Chapter 25. (We
will not be able to do so before then.)
 chapter 15

Event-causation and the root-problem

with CTM

There is an objection to our criticisms of CTM that is more important than any thus
far considered:
(FRM) Suppose that Smith (a human being) performs some computation. It is obvi-
ous that Smith’s behavior is not strictly morphology-driven. It is obvious that,
in general, paradigm-cases of computation differ in significant ways from the
so-called computational processes that, according to CTM, realize our cogni-
tive activity. As far as I can tell, you’ve only confirmed this triviality, and you
have therefore failed to make a case that CTM is false.
Let’s suppose that there is some operation –call it “computation*” – to be
defined thus. A computation* is a purely morphology-driven operation on bits
of matter. It is left open what, if any, semantic or otherwise representational
properties those bits of matter have. If we define CTM as the doctrine that think-
ing consists in computations*, all of your criticisms of CTM crumble. Those
criticisms all presuppose that the “computations” posited by CTM are like the
computations performed by engineers and mathematicians. And that presup-
position is false, given our new definition of CTM.
There are two points to make here. The advocate of (FRM) might be right to say that
my criticisms of CTM are null and void if one takes the computations posited by CTM
to be nothing more than bits of morphology colliding with other bits of morphology.
But given the points we’ve made in connection with terms “symbol”, “algorithm”, “syn-
tax”, “form”, and the like, it isn’t clear what intuitive motivation there would be for
(FRM). The idea behind CTM is that, in at least some cases, a cognitive achievement
results from strictly morphology-driven interactions among bits of matter. The ques-
tion is: why believe that semantically antiseptic interactions among dead bits of mor-
phology can constitute cognitive achievements? As it is traditionally defined, CTM
answers by saying that such interactions are relevantly similar to the cognitively preg-
nant symbolic-operations performed by logicians and mathematicians. But (FRM)
rejects this answer and doesn’t provide an alternative to it.
It could well be that, when somebody has a thought, that is in virtue of the fact that
some neural state of affairs has certain morphological properties. In fact, there is prob-
ably an element of truth in that view, given that our cognitive activity is almost cer-
tainly not entirely independent of the morphological properties of our brain-states. So
 Conceptual Atomism and the Computational Theory of Mind

(FRM) may be right. But (FRM) doesn’t have any meaningful similarity to (CTM), as
it is traditionally understood, since (FRM) isn’t similar to the view that thinking is
computing (at least not if, by “computing”, is meant anything even remotely similar to
what is conventionally meant by that word).
There is another difference between (CTM), as traditionally understood, and (FRM).
Supposing for the sake of argument that there is a one-one correspondence between
neural morphology and mental content, (CTM) aspires to explain that correspondence.
(CTM) aspires to explain how thought arises out of unthinking matter. If we grant
(CTM)’s supposition that strictly morphology-driven operations among bits of matter
are the essence of cognitive achievements like proof- and theorem-generation, it does
indeed become clear how it is that bits of dead matter can realize thought, and that is
obviously at least part of the reason that (CTM) is so widely accepted. But (FRM) explic-
itly refuses to endorse that very supposition, and it thus strips (CTM) of its raison d’être.
So even if it turns out that (FRM) is correct, that hurts (CTM) more than it helps it.

The root-problem with CTM

But there is a much more important point to make in response to (FRM). It can be shown
that, so far as interactions among brain-states (or physical objects any kind), constitute
instances of thought, those interactions must be driven by the representational (seman-
tic) properties of those states. This doesn’t mean that strictly morphology-driven interac-
tions don’t generate thought. But it means that, so far as they do, it is because representa-
tional properties are identical with, or realized by, morphological properties and,
consequently, morphology-driven interactions are ipso facto semantics-driven.
Let us start by recalling our point that it is not objects, but states of affairs, that
have causal properties and that figure into adequate causal explanations. It is not the
rock that breaks the window. What does so is the rock’s colliding with the window with
a certain amount of force at a certain instant. Language sometimes obscures this fun-
damental fact about causality. We say “Hitler was the cause of World War II.” But this
is obviously elliptical for some statement to the effect that such and such actions on
Hitler’s caused (or were partial causes of) World War II. Also, it is states of affairs, not
objects, that are caused. What is caused is not the statue, but is rather the existence in
a certain place and time of a piece of marble with certain properties. (In Book 2, Sec-
tion II, of the Physics, Aristotle writes that “Polyclitus [a famous sculptor] is the cause
of the statue”, and he also writes that “one can speak of the cause of a piece of bronze.”
But it is such and such acts on Polyclitus’ part, not Polyclitus per se, that bring about
the existence of the statue; and it is not the bronze per se, but rather the event of its
coming into existence, that is caused. Many of Aristotle’s views on causality, and in-
deed much post-Aristotelian thought on that topic, involve a failure to see that causal-
ity is a relation between states of affairs, not objects.)
 Chapter 15.  Event-causation and the root-problem with CTM 

I should point out that, in this context, the term “state of affairs” is meant to cover
both events as well as static conditions. Some artist is causally responsible for the fact
that there is a certain statue in a certain place and time. Here what was caused was a
(relatively) static condition. Of course, that condition was caused by way of events: the
artist had to chip away at a hunk of marble and then arrange for it to be moved to a
certain place. So static conditions are created by way of changes. In any case, both will
be referred to as “states of affairs.” (Incidentally, our usage of that expression is not
stipulative or neologistic, and is perfectly consistent with its existing meaning. As we
will see later, what we refer to as “static” states of affairs – e.g. a statue’s remaining in a
certain place – are uniform changes, and what we refer to as “events” – e.g. the statue’s
being moved from one place to another – are changes in the manner of change. So both
events and so-called static conditions are both changes, and our usage of the term “state
of affairs” in no way renders it ambiguous and thus doesn’t involve any equivocation.)1
A given object has many different causally effective properties. (Here the term
“causally effective” refers to any kind of causal potency, and is not being used in the
narrow technical sense that it had in Chapter 12.) Let R be a given rock. R has a certain
mass, a certain weight, a certain shape, a certain structure, a certain temperature, a
certain color, and so on. In virtue of having any given one of these properties, the rock
has certain causal powers that it wouldn’t otherwise have. Thus, these properties are not
causally inert. Of course, the rock does have some causally inert properties. For exam-
ple, it has the property of a thing x such that x is either a square or is not a square. But
the properties listed a moment ago are not in this category, and in this context we will
focus exclusively on the rock’s causally effective properties.
Not every property of the rock is involved in every causal transaction in which the
rock is involved. The fact that the rock has a temperature of 72° as opposed to 82° is
irrelevant to the fact that the window broke. This must be understood aright. Any two
non-simultaneous states have a causal connection with each other. So the rock’s having

1. Incidentally, this line of thought suggests a certain approach to an old metaphysical ques-
tion. It is often asked whether objects – rocks, trees, people, and so forth – have essential proper-
ties. Is there such a thing as the essence of the rock or the essence of your friend Smith? The
question is ill-formed, given that it is not objects, but states of affairs, that are causally and expla-
natorily important. When we talk about “essences” and “essential properties”, we are presumably
referring to things that have some kind of causal or explanatory significance. As we’ve seen,
Smith is a non-entity, as far as explanation and causation are concerned. What is not a non-en-
tity is Smith’s having a million dollars or Smith’s resenting you for your good looks. Obviously it
is essential to Smith’s having those properties that he have certain other properties. One does not
just have a million dollars. One’s having a million dollars supervenes on one’s having various
other properties – for example, one is in possession of bits of green paper that play a certain role
in the society of which one is a part. Since it is essential to Smith’s having a million dollars that
there exist some other, lower-level state of affairs of the kind just described, it seems to follow
that Smith’s having a million dollars has an essence. But it doesn’t follow that Smith has an es-
sence. And to the extent that essences are supposed to be explanatorily significant entities, it is
absurd to say that he, as opposed to his having this or that property, could have an essence.
 Conceptual Atomism and the Computational Theory of Mind

a temperature of exactly 72° has a causal connection with the shattering of the window.
But it doesn’t have any special or relevant connection: the window would have shat-
tered even if the rock had had a temperature of 82°. So even though it is causally effec-
tive in general, the rock’s having that temperature is inert in this particular context.
Of course, everything that we just said is true of brain-states. Let B be a particular
state (or structure or pattern of neural stimulation…) characteristic of Smith’s brain,
and let us suppose that B realizes Smith’s belief that two is a prime number. B has a
number of causally effective properties. It has thermal, electrical, and structural prop-
erties, and each of these is causally potent. In virtue of having any given one of these
properties, B has causal powers that it wouldn’t otherwise have. My own view is that B’s
representational properties belong on this list: in virtue of being a belief that 1+1=2, B
has causal properties that it wouldn’t otherwise have. But that is just what CTM denies,
and it is just what I am trying to establish; so it cannot be assumed in this context.
Not every case where B is involved in a certain causal transaction is a case where B’s
having a specific temperature is causally potent. In some contexts, B’s having that prop-
erty is potent, and in others it is not. For exactly similar reasons, supposing that B’s being
a belief that two is prime is causally potent, not every case where B is involved in a causal
transaction is a case where B’s having that particular property has any special causal role.
Some fiction will help us move onto the next phase of our argument. At time t Smith
suddenly thinks Socrates was wise. B is the brain-event realizing this mental event. Later,
at time t*, Smith suddenly thinks somebody was wise. B* is the brain-event realizing this
mental event. Further, B is what causes B* to come into existence. Of course, given what
we just said about causality, what this really means is: some state of affairs involving B is
causally responsible (at least in part) for some state of affairs involving B*.
Given only what we’ve said, does it follow that Smith’s thinking Socrates was wise
had anything to do with his later thinking somebody was wise? No. Given only that B is
what caused B* – more exactly, given only that some state of affairs involving B is what
led to there being some state of affairs involving B* – it doesn’t follow that Smith’s think-
ing Socrates was wise had any special connection with his subsequently thinking some-
body was wise. That would follow only if B’s being a belief that Socrates was wise is what
led to the state of affairs consisting in B*’s being a belief that somebody was wise. If it was
B’s having a certain temperature that was responsible, and not its being a belief that
Socrates was wise, then Smith’s thinking Socrates was wise at t didn’t have anything to
do with Smith’s thinking somebody was wise at t*. We thus cannot conclude that Smith
inferred that somebody was wise on the basis of his belief that Socrates was wise. So we
cannot conclude that Smith thought: Socrates was wise, therefore somebody was wise.
Obviously Smith’s thinking Socrates was wise constitutes a cognitive achievement,
and his thinking somebody was wise constitutes another cognitive achievement. But it
doesn’t follow that his thinking both of those thoughts constitutes some third cognitive
achievement. For this to follow, it would be necessary (though not sufficient, as we will
see) that his thinking Socrates was wise was responsible for his later thinking somebody
was wise. If two thoughts are causally unrelated to each other, then surely they don’t
 Chapter 15.  Event-causation and the root-problem with CTM 

jointly constitute some third thought. We’ve seen that Smith’s thinking Socrates was
wise doesn’t necessarily have any (relevant) causal connection to his later thinking
somebody was wise. So we cannot conclude that his thinking both of those constitutes
some third cognitive achievement. For example, we cannot conclude that it involves
his making an inference. It is irrelevant that B is what caused B*, and it is irrelevant that
the proposition encoded in B logically entails the proposition encoded in B*.
We can reinforce this point by extending our story. Later that day, after reading an
article about some historical figure (other than Socrates) who committed an unwise act,
Smith suddenly thinks somebody was not wise. This happens at time t**. It would be
absurd to say that, in virtue of thinking Socrates was wise at t, and then thinking some-
body was not wise at t**, Smith was guilty of a logical non-sequitur. In fact, it would be
absurd to say that, in virtue of having those two thoughts, Smith deserved credit for
having any thought-process that embodied any view at all as to how, if at all, those two
propositions are related. The reason this would be absurd is obviously that his thinking
Socrates was wise didn’t have anything to do with his thinking somebody was not wise.
(In any case, those two events didn’t have any special or relevant causal connection,
even though, like all non-simultaneous events, they have a causal connection.) For
similar reasons, it would be absurd to say that, in virtue of thinking Socrates was wise at
t and somebody was wise at t*, Smith has had a thought that embodies any view at all as
to how those propositions are related to each other. It is therefore out of the question to
say that Smith deserves credit for having made a logical inference.
The points just made can be generalized in a few different directions. Let us start
with the most obvious direction of generalization. Let P be and P* be any two proposi-
tions such that P logically entails P*. Further, let bp and bp* be two brain-states, both
had by a single person X, such that bp realizes a belief that P and such that bp* realizes
a belief that P*. Supposing that bp causes bp*, it doesn’t follow that bp’s being a belief
that P has anything to do with bp*’s coming into existence. It could be that it was bp’s
having some non-semantic property that was thus responsible. It therefore doesn’t fol-
low that X’s thinking P has anything to do with his later thinking P*. There is no reason
to believe that X inferred P* from P or, indeed, that he had any thought-process that
embodied any views, whether meritorious or not, as to how those two propositions
might be related.
Let us further generalize our analysis. Let T and T* be any two non-simultaneous
events or objects that are representational. For the sequence consisting of those events
to form a single representation, it is not enough that T be what causes T* to come into
existence: it is necessary that T’s being a representation be what does so. A bit of fiction
may illustrate my meaning. At 3:00 p.m. I am feeling very sad. I attempt to write “I am
suffering from great sadness.” I manage to write “I am suffering from great.” But right
after I write the “t” in “great”, I am struck by the beauty of my own penmanship, and
this causes me to feel great joy. Because of this, I forget about my earlier intention to
write “I am suffering from great sadness”, and I form the intention to write “sadness
 Conceptual Atomism and the Computational Theory of Mind

never lasts and, ultimately, life is a great joy.” But right after I write the second “s” in
“sadness”, I am interrupted by the doorbell. I never write in my diary after that.
Under these circumstances, have I produced a token of “I am suffering from great sad-
ness”? No. I have only produced an ink-mark that looks just like a token of that sentence.
To be sure, the token that I produced of “I am suffering from great” is causally responsible,
at least in part, for the adjacent token of “sadness.” Nonetheless, I never managed to carry
out my original intention of tokening the sentence “I am suffering from great sadness.”
Therefore, that sentence was not tokened in this context, even though the ink-mark I have
produced is morphologically just like a token of that sentence.
Let T be my inscription of “I am suffering from great”, and let T* be my inscription
of “sadness.” It was T’s having certain aesthetic properties, and not its having a certain
meaning, that led to my writing T*. So T’s being a symbol (or, in any case, T’s being a
symbol that has a certain meaning) had nothing to do with T*’s coming into existence.
So far as they have semantic properties, the inscriptions “I am suffering from great” and
“sadness” are causally unrelated (except in the trivial sense in which any two non-si-
multaneous entities are causally related), and thus have nothing to do with each other,
that being why they don’t jointly form a sentence-token.
Supposing that X and Y are symbol-tokens, the juxtaposition of X and Y doesn’t
constitute a single symbol-token, unless X’s being a symbol (or representation) is what
generates Y (or vice versa). If this condition isn’t met, then X has no semantically sig-
nificant relation to Y, in which case it immediately follows that they don’t jointly form
any semantic unit. The juxtaposition of these symbols may look (or sound) like a
meaningful unit, and the causal mechanism that mediates between X and Y may be
completely reliable. But if it isn’t X’s being a representation that generates Y (or vice
versa), then X and Y aren’t parts of a single representation. Where semantics is caus-
ally inert, representations cannot combine to form more complex representations, and
nothing has any representationally significant relation to anything else.
Some thoughts consist of other thoughts. If I think grass is green and snow is white,
that thought involves the thought grass is green and also the thought snow is white.
CTM says that semantics is causally inert. Given what we just saw, CTM thus cannot
accommodate the fact that there are thoughts that consist of other thoughts. CTM is
therefore inconsistent with the fact that there are complex thoughts.

Necessary versus sufficient conditions for the formation of complex thoughts

Suppose that I think Socrates was wise and then later think somebody was wise. Let b
and b* be the brain-states mediating those thoughts. As we just saw, if b and b* are to
constitute a single thought, or are to be part of the same thought-process, it is necessary
that b’s being a thought that Socrates was wise be at least part of what generates b*. But
it is not sufficient, as the following story shows.
 Chapter 15.  Event-causation and the root-problem with CTM 

A hypnotist programs me to think something was wise whenever my heart rate goes
above 90 beats per minute. One day it suddenly occurs to me that Socrates was wise. (I
finally understand some obscure passage of The Republic, and I see that, contrary to
what I used to think, Socrates really was wise.) The resulting euphoria causes my heart-
rate to sky-rocket. Because of my hypnotic programming, this causes me to think some-
thing was wise. Here we have a case where my thinking Socrates was wise causes me to
think something was wise. So if b and b* are the brain-states associated with those two
thoughts, we have a case where b’s being a thought that Socrates was wise is responsible
for b*’s being a thought that something was wise. But we don’t have a case where I have
thought Socrates was wise, therefore something was wise and, more generally, we don’t
have a case where a single thought or thought-process comprises both of those thoughts.
An exact analogue of what we just said about b and b* could be constructed in connec-
tion with ink-marks or noises or any other symbol-tokens.
In general, for the juxtaposition of two representational entities (be they thoughts
or symbol-tokens) to constitute a single, complex representation, it is necessary but not
sufficient that the one’s being a representation of a certain kind be causally responsible
for the generation of the other.
Of course, it is an interesting question what other conditions must be fulfilled. But in
this context we don’t need to know the answer to that question. CTM demands that se-
mantics be totally inert, and it thus renders impossible the fulfillment of one of the condi-
tions necessary for the combining of representations into other representations. This by
itself is a major strike against CTM, given how frequently such combinations occur.

Semantic versus conceptual complexity

To develop this line of thought, and to see more fully why CTM must be rejected in
light of it, we must first distinguish between two different kinds of complexity. Con-
sider an inscription of the word “Socrates.” That inscription is physically complex,
since it consists of a multiplicity of marks. But it is not semantically complex.2
Any brain-state that realizes a concept of Socrates will inevitably have a great deal
of physical complexity. But it doesn’t follow that any such state has representational
complexity. I myself believe that anything that realizes a concept of Socrates will have
representational complexity. (We argued for this in Chapter 1.) But even if anything
that realizes a concept does have representational complexity, we must distinguish be-
tween the kind of complexity that such a thing has in virtue of being such a concept
and the kind of complexity that it has for some other reason.

2. Strictly speaking, it isn’t quite right to say without qualification of that inscription that it is
physically complex. Whether it is complex, and if so how, is a function of the explanatory
context. But we needn’t pursue this.
 Conceptual Atomism and the Computational Theory of Mind

With this distinction in place, let us tell another story. At time t, Smith thinks
Socrates. b is the brain state realizing this thought. At some later time t*, Smith thinks
snored. b* is the brain-state realizing this thought. For exact analogues of the reasons
just given, unless b’s being a representation of Socrates is responsible for b*’s subse-
quent occurrence, Smith has not thought Socrates snored, at least not in virtue of the
fact that b and b* occurred. Semantics cannot be inert if anyone is to think anything
true or false. No true or false thought is representationally simple. So nobody can think
anything true or false except in so far as it is one’s brain-states’ having representational
properties that are governing one’s cognitive activity. It follows that CTM strips anyone
of the ability to think anything true or false, given that, according to CTM, it is never
in virtue of its semantic properties that a brain-state has causal powers.
I myself believe that, contrary to what the argument just given seems to presup-
pose, one cannot just think Socrates or snored. Supposing that this is right, it might
seem that my argument crumbles. But it does not crumble, and this is because, as we
discussed earlier, CTM needs it to be possible to think just Socrates or snored or blue.
CTM demands that we have atomic concepts, and that our thoughts consist of con-
catenations of these concepts. So our argument shows that CTM fails relative to its
own assumptions.
CTM requires the truth of some kind of conceptual atomism. So conceptual atom-
ism would indeed be de rigueur if CTM were correct. But we’ve seen reason to think that
CTM is not correct. It involves misunderstandings of the nature of symbolism, formal
truth, and causation; and it doesn’t appear that any doctrine significantly similar to CTM
would not involve these misunderstandings. So there is no reason to accept conceptual
atomism, given only that CTM demands its acceptance. In the next three chapters, we
will see why some other reasons to accept conceptual atomism are in fact spurious.
 chapter 16

Fodor’s first argument for conceptual atomism

Fodor accepts CTM. He also sees that CTM demands the truth of conceptual atomism
(henceforth “atomism”), and this is at least part of the reason that he accepts atomism.
But Fodor realizes that few are willing to accept atomism, and he thus provides three
independent arguments for it (Fodor 1998). In this chapter, we will examine the first of
these three arguments. We will examine the other two in Chapters 17 and 18.

Fodor’s argument

There is no doubt that some expressions are definable. “Triangle” can be defined as
“closed, straight-edged, planar figure such that any two, but not all three, of its sides
intersect.” “Two” can be defined as “the successor of one.” In general, expressions
belonging to “bona fide axiomatic systems” can be defined. Further, “patently
phrasal” expressions can be defined, at least in a relative sense: “smart brown cow”
can be defined as “anything x such that x is smart, x is brown, and x is a cow.”
But leaving aside these two classes of expressions, virtually no expression can
be defined. No one has produced an adequate definition of “knowledge.” (It used
to be thought that “knowledge” could be defined as “justified true belief.” But Get-
tier 1963 famously proved this false, and attempts to produce an adequate defini-
tion have failed. See Shope 1983.) What we just said about “knowledge” is true of
practically any expression denoting a philosophically, or otherwise theoretically,
important concept. In fact – what is probably more important – even expressions
denoting pedestrian concepts turn out to be incapable of definition. Consider the
verb (not the noun) “paint.” Contrary to what might initially be thought, this can-
not be defined as “to cover in paint” or even “to intentionally cover in paint.”
When you dip your paintbrush into a bucket of paint, you are not painting it, even
though you are intentionally covering it in paint. There are other prima facie rea-
sonable definitions of this verb, but they all fail. Given that the word “paint” is
indefinable, it follows that the corresponding concept is simple. After all, if that
concept were complex – if it consisted of other concepts (e.g. cover, intentional,
and so on) – then there would be some true statement saying how that concept
were composed of those other concepts (just as there is a true statement saying
how water molecules are composed of hydrogen and oxygen molecules). There
isn’t; so it doesn’t. “Paint” is indefinable and the concept that it expresses is there-
fore simple. Given virtually any other expression, what we just said about “paint”
is true of that other expression, showing that all concepts are simple. (In any case,
 Conceptual Atomism and the Computational Theory of Mind

it shows that all, or nearly all, the concepts denoted by semantically non-complex
expressions of natural language, leaving aside those concepts that are expressed by
artifacts like “integral”, are simple.)1

Our response

In this context, the word “concept” obviously doesn’t denote a psychological entity. So
if cogent, the argument just presented shows that concepts in the non-psychological
sense of the word are simples. That argument does not, at least not directly, say anything
about concepts in the psychological sense. To simplify further discussion, let us refer
to concepts in the non-psychological sense as “conceptso”, and let us refer to concepts
in the psychological sense as “concepts.” So a concept is one’s grasp of something, and
a concepto is a platonic entity. (The subscript stands for “objective.”)
Fodor is trying to show that concepts (no subscript) are atoms (simples). He is trying
to make a case for conceptual atomism. So Fodor must be assuming that if conceptso are
simples, the same is true of our psychological representations of them. In any case, if this
assumption is not granted, then Fodor’s argument doesn’t show that concepts are atoms.
So for the sake of argument, let us grant Fodor’s tacit assumption that conceptual atom-
ism follows from conceptualo atomism. (Later we will find this assumption to be false.)
Granting this assumption, there are two major problems with Fodor’s argument.
Nobody denies that x’s being a belief is necessary for x’s being knowledge, and nobody
denies that the necessity in question is analytic, and not empirical, in nature. Nobody
denies that putting paint on something is necessary for painting it, and nobody denies
that the necessity in question is analytic. Given any concepto C, nothing is easier than
to identify conditions that are analytically necessary for something’s falling under it.
(Actually, Quine and a few others do deny that if x is knowledge, x is a case of belief is
analytic. At the end of the present chapter, we will examine their reasons for holding
this. For now we will operate on the reasonable – and, we will see, demonstrably cor-
rect – view that many propositions are analytic.)
But surely analytic necessities correspond to conceptualo structure. x is an instance
of knowledge does not entail x is an instance of putting paint on something. But the latter
is entailed by x is a case of painting. Surely this tells us that the architecture of the one
concepto is different from the architecture of the other.
In fact, this last point would seem to be a truism: given that to fall under the one
concepto but not the other, x must be an instance of belief, those two conceptso are ipso
facto structurally different. When we talk about the “structures” (or “compositions” or
“architectures”) of conceptso, we obviously aren’t referring to their physical structures.
We are referring to relations of logical dependency, not of spatiotemporal juxtaposi-
tion. With a minor qualification to be stated shortly, for conceptso C and C* to differ in
respect of their structures just is for x is a C to entail something not entailed by x is a

1. Fodor (1998: Chapter 4).

 Chapter 16.  Fodor’s first argument for conceptual atomism 

C*. A corollary is that, as long as we can give necessary (and non-trivial) conditions for
x’s falling under C, it follows that C has conceptualo structure. (A trivial condition
would be given by a statement like: x falls under C iff x falls under C.)
Taken in conjunction with this line of thought, Fodor’s own argument would seem
to show that any given concepto has an infinite amount of structure. Getteriologists
may not have provided necessary and sufficient conditions for something’s being
knowledge. But they have succeeded in providing a number of necessary conditions
(Shope 1983). Given what we said a moment ago, each of these new necessary condi-
tions corresponds to a newly discovered fact about the architecture of the concepto of
knowledge. If there are infinitely many such conditions, then that concepto has an infi-
nite amount of structure. Supposing that there are only finitely many such conditions,
that concepto has a finite, but non-null, amount of structure and, more importantly,
that concepto is definable. Either way, Fodor’s argument collapses.2

2. Let us address a possible objection to our argument:

“Not all analytic dependencies correspond to conceptualo structure. x is knowledge entails x
is not a square circle. But this analytic dependency doesn’t show us anything specifically about
the concepto of knowledge. After all, x is a case of painting entails x is not a square circle, as do x
is a house and x is a barber. The fact that x is knowledge entails x is not a square circle evidently
tells us nothing about the structure of the concepto of knowledge. So it would appear that you
were wrong to see analytic dependencies as indicating conceptualo structure, and that your ar-
gument against Fodor was misguided.”
This objection actually reinforces our criticism of Fodor’s argument. Consider the entail-
ment given by the statement:
(*) “For any x, if x is knowledge, then x is belief.”
If the occurrence of “knowledge” is replaced with an occurrence of “a house” or “case of pain-
ting”, what results is a falsehood. So (*) does give us differential information about the structure
of the concepto of knowledge, thus confirming our point. It is an interesting question what exact-
ly the difference is between (*), on the one hand and:
(**) for any x, if x is knowledge, then x is not a square circle,
on the other. More generally, it is an interesting question what precisely the difference is between
those analyticities that reveal conceptualo structure and those that do not. But it is not a question
that has to be answered here. For our purposes, it is enough to know that analyticities typically
(though not universally) reveal conceptualo structure.
But for what it’s worth, it seems to me that the difference between (*) and (**) is that (**) is
semantically true, whereas (*) is not. As we saw in Chapter 13, if a sentence is semantically true,
that is because of structural properties of its derivation tree, a consequence being that its truth
has nothing to do with the specific contents, or therefore the specific conceptso, meant by the
expressions constituting that sentence. Consequently, such a sentence doesn’t expose what is
distinctive about those conceptso and thus doesn’t provide any information about them. By the
same token, if a sentence is non-semantically but analytically true, its truth does turn on the
peculiarities of the meanings, and therefore the specific conceptso, assigned to its constituent-
expressions. Consequently, such a sentence does expose what is distinctive about those concepts,
and it therefore does provide information regarding the structures of those conceptso.
 Conceptual Atomism and the Computational Theory of Mind

Here is a useful analogy. It is a point often made by philosophers of science that the
term “explanation” is contextual.3 To provide a complete explanation of anything, I
would have to describe every event that led up to that event, starting with the Big Bang.
In this sense of the word “explanation”, it is impossible in practice, and probably in
principle, to explain anything. But it is a datum that perfectly good explanations are
given by statements like “Smith crashed the car because he was drunk”, “Jones is limp-
ing because he twisted his ankle while playing squash”, and so on. So given only that
“complete” explanations cannot be given, it would be absurd to conclude that explana-
tion tout court is impossible. It would be absurd to say: “the world is explanatorily
amorphous, since nothing can be completely explained.”
Similarly, it would be absurd to say: “the concepto of knowledge is explanatorily
amorphous, since there is no complete account of its analytic structure.” Given that
there are partial characterizations of its structure, it follows that it is not analytically
amorphous.

An objection to our assessment of Fodor’s argument

There is an objection to this line of thought that we must consider.4 And it is not just
an objection: it is one that embodies a position that Fodor himself holds and that un-
derlies his various arguments for conceptual atomism:
According to the argument you just gave, C and C* have different structures ex-
actly if they differ in their entailment-relations. With a few innocuous qualifica-
tions, statement:
(S) “x is identical with Socrates”
doesn’t entail anything not also entailed by:
(A) “x is identical with Aristotle.”
We know from Kripke that the sentence “Socrates was bald” is not analytic. We
also know from Kripke that, if we replace the occurrence of “Socrates” in that
sentence with any other proper name the result will not be analytic. Thus, “Plato
was bald” is not analytic (even though it is true), and neither is “Napoleon was
bald.” In general, uniformly replacing proper names with other proper names
never turns a non-analytic truth into an analytic truth or (for similar reasons) vice
versa. This suggests that the concepto expressed by the word “Socrates” – this be-
ing the concepto identical with Socrates (or some such) – has no analytic structure
not also had by the concepto expressed by the words “Plato” or “Napoleon.” Your

3. See, for example, Carnap (1966).

4. It is similar to an objection that George Rey (1996) makes of Peacocke’s (1992, 1996) analysis
of conception. We will discuss both Peacocke’s analysis, and Rey’s objection to it, in Chapter 24.
 Chapter 16.  Fodor’s first argument for conceptual atomism 

position is that conceptso differ just in case their analytic liaisons differ. If you are
right, then the concepto Aristotle (or identical with Aristotle) is structurally just
like the concepto Plato. By itself, this doesn’t mean that those conceptso have no
structure. (It could be that x is identical with Aristotle entails x is a spatiotemporal
object.) But it does show that, so far as those conceptso differ, that has nothing to
do with structural differences between them. That, in its turn, shows that grasping
the one concepto as opposed to the other cannot be understood in terms of grasp-
ing structural differences between them, and thus cannot be understood in terms
of one’s grasping some set of conceptso C1…Cn such that the one concept, but not
the other, is constructed out of those conceptso. So one’s grasping the concepto
Plato as opposed to the concepto Aristotle is not to be understood in terms of what
other conceptso one grasps. That difference is conceptually primitive. What we
just said about the conceptso Plato and Aristotle is true of countless other pairs of
conceptso. Indeed, what we just said holds of practically any two conceptso. The
pairs of conceptso that you chose seem to be more the exception than the rule.
What you said about the pair of conceptso knowledge and paint seems not to apply
universally or even usually. So, given only what you’ve said, Fodor’s argument may
go through, at least for the great majority of conceptso. Further, it seems clear that
differences between conceptso are often primitive, i.e. if C and C* are different
conceptso, that fact cannot typically be understood in terms of how those con-
ceptso are composed out of other conceptso. So it would seem that, in the end,
Fodor’s argument does go through, notwithstanding some of the technical prob-
lems that you have pointed out with it.
We have spoken a great deal about the difference between, on the one hand, what is
literally meant by expressions and, on the other hand, what we learn in the process of
ascertaining what is literally meant by them. The objection just stated involves a failure
to appreciate this difference. As we’ve seen, there is some x such that what is literally
meant by any token T of:
(HL) “Hesperus is lovely”,
and also by any token T* of
(PL) “Phosphorous is lovely”,
is simply:
(XL) x is lovely.
But one must compute the meanings of T and T* on the basis of background semantic
and pre-semantic knowledge; and for this reason, those two tokens may diverge dra-
matically in respect of what they impart to one. For a moment, let us discuss the con-
ceptso Hesperus and Phosphorous (or identical with Hesperus and identical with Phos-
phorous). Given the points that we will make in connection with these conceptso, it will
be clear what we must say about the conceptso Socrates and Plato.
 Conceptual Atomism and the Computational Theory of Mind

First of all, which entity are we talking about when we talk about “the concepto
Hesperus” or “the concepto identical with Hesperus”? If we are talking about the infor-
mation through which one computes the meanings of sentence-tokens like T, then that
concepto is replete with conceptualo structure. As we saw in Chapter 4, the information
that one associates with T (in our technical sense of “associates with”) is some exist-
ence-claim having the form:
(EHL) There is some luminescent beautiful x in such and such part of the morning
sky…and what is meant by “Hesperus is lovely” is: x is lovely.
So on one possible disambiguation, the term “the concepto identical with Hesperus”
refers to this wide-scope descriptive information.5
What else might the expression “the concepto of Hesperus” refer to? Hesperus’ exist-
ence is not a simple, ultimate metaphysical fact. Given any possible world W where Hes-
perus exists, its existence supervenes on various other facts. Certain mass-energy dis-
placements, having certain origins, resulted in a large solid body having a certain
trajectory. Of course, Hesperus is not modally frozen: there are possible worlds where,
because of some cataclysm, it breaks free of its orbit around the sun. More generally, there
are worlds where it has properties very different from those that it actually has. But there
are no worlds where Hesperus has totally different origins from those that it actually has.
Given this last point, it would seem that any world W where Hesperus exists is one
that has much in common with our world apart from its having the property of includ-
ing Hesperus: some space-time region R of W must comprise innumerable sub-atomic
events that are qualitatively much like the events comprised by some corresponding re-
gion in our world. Further, because Hesperus is presumably individuated by its origins,
it exists in a world only if that world comprises at least some objects (other than Hespe-
rus) that are numerically identical with objects in our worlds, the reason being that a

5. Or, more likely, it refers to some kind of idealization of this information. As we discussed,
the information through which one computes the meaning of tokens of “Hesperus”, and thus of
sentence-tokens containing that term, varies from person to person. No two people will learn
what is meant by “Hesperus” in quite the same way. Given any two people to whom that term is
defined ostensively, their respective optical, and subsequent epistemic, relations to Hesperus will
probably not be exactly the same; and the same is true of any two people to whom it is defined
indirectly or non-ostensively. So far as there is one “concepto of Hesperus” – i.e. so far as the term
“the concepto of Hesperus” is not an improper, and thus non-referring, definite description –
that term seems to refer to some concepto that abstracts from all of these person-to-person va-
riations, leaving only some core concepto like unique last body to disappear from the morning sky,
or some such. (Of course, this is exactly what Frege and Russell, and indeed most pre-Kripke
semanticists, would have identified as “the concepto of Hesperus.”)
What is important here is that if by “the concepto of Hesperus” one means the information
through which people think about Hesperus – or, more exactly, through which they compute the
semantics of occurrences of expressions like T – then the concepto of Hesperus is by no means
conceptuallyo simple, as it consists of the conceptso morning, luminescent, and so on.
 Chapter 16.  Fodor’s first argument for conceptual atomism 

world that comprised none of the same micro-particles as our world would ipso facto fail
tom comprise anything that had the same origins as Hesperus. (And, of course, the same
thing mutatis mutandis holds of those other objects – of those various micro-particles.)
Further, if a world comprises those other objects, and those objects are arranged in the
right way, then that world ipso facto comprises Hesperus. So there is some set of proper-
ties P1…Pn such that, for a world W to comprise Hesperus, it is necessary and sufficient,
that each of these properties be instantiated in W, and such that none of these properties
will be identical with Hesperus or identical either with Hesperus or with a square circle or
any other property that must itself be understood in terms of the concepto of Hesperus.
It seems to me that, on one possible disambiguation, the term “the concepto of
Hesperus” refers to this set of properties: the set of properties such that their joint in-
stantiation is necessary (and sufficient) for Hesperus’ existence in any possible world
W. Relative to this disambiguation, the concepto of Hesperus is replete with concep-
tualo structure. For exactly similar reasons, “the concepto of Socrates” would refer to a
conceptuallyo complex entity. And this, of course, would scuttle the objector’s point
that “the concepto of Hesperus” and “the concepto of Socrates” refer to conceptuallyo
simple entities.
Inevitably, some will respond by saying:
I have a perfectly good grasp of the conceptso Socrates, Hesperus, and Aristotle. But
I haven’t any knowledge as to the nature or identity of the sub-atomic events in
virtue of whose occurrences these conceptso are instantiated. So it is not really an
option to say that the concepto of being identical with Socrates is to be understood
in the way you just proposed.

Supposing that this is right, what does it mean to say that one “grasps the concepto of
being identical with Socrates”? It means that one is able to think about Socrates. Of
course, to think about Socrates, it is not necessary to know much at all about his ori-
gins, let alone about their molecular basis. But as we’ve seen, to think about Socrates,
one does have to know some rather complex description that Socrates alone uniquely
satisfies. The objector is quite right that, in this sense of the term “concepto of Socrates”,
one doesn’t have to know anything about the molecular basis of Socrates’ conditions of
origination to grasp a concepto of Socrates. But this is consistent with our point that
anything fit to be denoted by the term “the concepto of Socrates” has conceptualo struc-
ture. Of course, what we just said about “Socrates” is true of “Aristotle”, “Hesperus”, and
any other singular term. This shows that the thing denoted by “the concepto of Hespe-
rus”, on either disambiguation of that expression, has a structure very different from
the thing denoted by “the concepto of Socrates.”
Of course, the objector is entirely right to say that that, in terms of their literal
meanings, “Socrates” and “Aristotle” are equally devoid of descriptive content. And he
is right to say that, in terms of their literal meanings, FAristotle has phiG will be ana-
lytic iff the same is true of FPlato has phi.G (This last statement assumes, of course, that
phi is not a predicate like “identical with Aristotle” or “identical with Plato.”)
 Conceptual Atomism and the Computational Theory of Mind

But as we’ve seen (Chapters 1–4), nothing at all can be inferred from this fact con-
cerning the structure of the concepto of Aristotle or of the concepto of Socrates. An
exact analogue of the argument just given shows that we just said about singular terms
is also true of terms denoting natural kinds (“water”, “wood”).

Does conceptual atomism follow from conceptualo atomism?

As we noted earlier, Fodor’s argument tacitly assumes that the truth of conceptualo at-
omism is sufficient for the truth of conceptual atomism. Is this assumption correct?
No. While a case can be made that conceptualo atomism is necessary for conceptual
atomism, no case can be made that it is sufficient.
For the sake of argument, let us suppose that (with the obvious exceptions that
Fodor discusses) conceptualo atomism is correct, and let C and C* be two conceptso.
Given only that these conceptso are simple, i.e. that no conceptso (other than them-
selves) are constitutive of them, it doesn’t follow that my mental representations of
them are conceptually simple. Let X and Y be two qualitatively identical, perfectly
homogeneous spheres that are located in different places. I think of X as the sphere in
Aunt Jenny’s basement and I think of Y as the sphere in Uncle Fred’s attic. So even though
X and Y are qualitatively identical, and even though they are both lacking in internal
structure, it doesn’t follow that my concepts of them are qualitatively identical or that
my concepts of them are without internal articulations.
Given only that two different conceptso have no structure, and thus have the same
structure, it doesn’t follow that my concept of the one is qualitatively identical with my
concept of the other. What follows is that I cannot distinguish those conceptso on the
basis of their internal structure.
Of course, X and Y are spatiotemporal individuals, whereas conceptso are not; and,
conceivably, one might therefore argue that where non-spatiotemporal entities are
concerned, the structures of our concepts do mirror those of their objects. But Fodor
provides no argument for such a view and, more importantly, such a view would be
extremely implausible. Structurally identical spatiotemporal individuals can be distin-
guished by their differences in spatiotemporal location. But structurally identical con-
ceptso could not be distinguished in this way. So far as conceptso can be distinguished
from one another, it must therefore be on the basis of their structures. In any case,
Fodor’s tacit assumption (viz. that conceptualo atomism is necessary for conceptual
atomism) is false.
 Chapter 16.  Fodor’s first argument for conceptual atomism 

Quine on analyticity

This is a good place to discuss Quine’s (1951) analysis of the analytic-synthetic distinc-
tion. Quine holds that, with a few trivial exceptions, all sentences are synthetic. (The
exceptions involve cases where one term has been stipulated to have the same meaning
as another.) Our argument against Fodor assumes that there are analytic truths. In-
deed, we have made liberal use of this assumption in this work, and we must therefore
consider Quine’s views on analyticity. It should be pointed out that Fodor (1998: 25,
45–46) agrees with Quine. Indeed, Fodor takes it as a foregone conclusion that Quine’s
arguments are cogent, give or take some nuances.
Here is the basic structure of Quine’s argument. Analyticity is to be understood in
terms of synonymy. But synonymy must be understood in terms of analyticity. Therefore
the concept of analytic truth is a viciously circular and therefore incoherent one. Let us now
state Quine’s argument in full.
If indeed “bachelors are unmarried males” is analytic, that is because “bachelor” is
synonymous with “unmarried male.” Similarly, if “triangles are three-sided, closed,
planar, straight-edged figures”, that is because “triangle” is synonymous with “three-
sided, closed, planar, straight-edged figure.” In general, analyticity is to be understood
in terms of synonymy.
But what is it for two expressions to be synonymous? It is not merely for all intersub-
stitutions of them to preserve truth-value. “Rhenates” and “chordates” can be intersub-
stituted salva veritate, but they are obviously not synonymous. For two expressions to be
synonymous, intersubstitutions of them must preserve not only truth-value, but also
meaning. But the concept of meaning is itself to be understood in terms of the concept
of analyticity: Fx has phiG has the same meaning as Fx has psiG only if Fx has phi iff has psiG
is analytic. So analyticity is to be defined in terms of synonymy, and synonymy is to be
defined in terms of analyticity. Attempts to define analyticity are thus doomed to failure,
since they are necessarily characterized by the same vicious circularity as attempts to
inductively justify induction, and the concept of analyticity is therefore incoherent.
Having presented the preceding line of thought, Quine correctly notes that an
objection can be made to it. Strictly speaking, “rhenates” and “chordates” cannot al-
ways be intersubstituted salva veritate. Even if “Smith believes that rhenates are chor-
dates” is false, “Smith believes that rhenates are rhenates” may be true. In general, in-
tensional contexts don’t always tolerate intersubstitutions of non-synonymous
co-referring (or co-extensive) terms. For this reason, contrary to what was alleged a
moment ago, “synonymous” can be defined; for it can be defined as “intersubstitutable
salva veritate in intensional contexts.” And this breaks the definitional circle.
Quine has no trouble rebutting this point. The concepto of an “intensional” context
is itself to be understood in terms of the concepto of meaning (and, therefore, in terms
of the concepto of analyticity). C is an intensional context iff replacing any term occur-
ring in it with a co-referring term is guaranteed to preserve the truth-value of the host-
sentence only if that intersubstitution involves expressions having the same meaning.
 Conceptual Atomism and the Computational Theory of Mind

As we saw a moment ago, the concepto of meaning is to be understood in terms of the

concepto of analyticity. So “synonymous” cannot be non-circularly defined as “inter-
substitutable salva veritate in intensional contexts.”

Evaluating Quine’s argument

If there are any analytic sentences at all, one of them is:

(*) “n is a unique even prime iff n is one less than three.”
Quine explicitly assumes that the expressions flanking the biconditional must be syn-
onymous if that sentence is to be analytic. But the expression “n is a unique even prime”
obviously isn’t synonymous with the expression “n is one less than three.” If those two
expresions were synonymous, i.e. if they had precisely the same linguistic meanings,
then (*) would be as trivial as “a yard is a distance of three feet.” Analytic truth isn’t
always definitional truth (that being why analytic truth isn’t always trivial). The sen-
tence “if an event begins at 3:00 p.m., then it doesn’t begin at 3:02 p.m.” is analytic but
not definitionally true, since (for well-known reasons) “begins at 3:00 p.m” is not syn-
onymous, or otherwise equivalent, with “doesn’t begin at 3:01 p.m. or at 3:02 p.m. or...”
or with any other comparable expression. Quine is therefore wrong to assume that
there is analyticity only where there is synonymy, and his argument thus implodes.
Conceivably, Quine might deny that (*) is analytic on the grounds that the expres-
sions flanking the biconditional are not synonymous. But in that case, he would simply
be redefining the term “analytic”, and his argument would have force only against an
artifact of his own definitions.
If we are to make this argument more precise, we must note that the term “analytic”
is ambiguous. Sometimes when a sentence is described as “analytic”, what is meant is
that it encodes a conceptuallyo true proposition: a proposition that holds entirely in
virtue of the structures of the conceptso composing it. An example of a sentence that is
analytic in this sense would be: “there is no law where there is no government.” Surely
it is not an empirical fact that there is law only where there is government, it being part
of the concepto of law that instances of it presuppose the presence of government.
But sometimes when a sentence is described as “analytic”, what is meant is that it
is formally true. Even though no empirical knowledge, above such as is needed to grasp
its meaning, is needed to determine the truth-value of “there is no law where there is
no government”, that sentence is not formally true. By contrast, a token of “I am here
now” is formally true, as is a token of “if snow is white, then snow is either white or
green.” As we discussed earlier, a sentence-token T belonging to language L is formally
true iff it is a theorem of the semantic rules of L that T is true.
Quine’s statement that analyticity is to be understood in terms of synonymy is
therefore ambiguous. Let us consider each disambiguation of Quine’s statement.
 Chapter 16.  Fodor’s first argument for conceptual atomism 

The concepto of formal truth is not, at least not on the face of it, to be defined in terms
of that of synonymy. (A sentence or sentence-token T is formally true if its truth is a theo-
rem of the relevant semantic rules. There is no mention of synonymy here.) So Quine’s
argument implodes if by an “analytic” sentence, Quine means one that is formally true.
To be sure, a defender of Quine’s position has what initially appears to be a compel-
ling response to this. If the concepto of a semantic theorem, and thus of a formal truth,
is to be understood in terms of the concepto of analyticity, then in defining an “analytic”
truth as one that is formally true, one has defined analytic truth in terms of itself. This,
of course, would vindicate Quine’s view that the concepto of analyticity is a viciously
circular and therefore incoherent one. And, at first, it does seem that the concepto of
logical truth is to be understood in terms of the concepto of analyticity. After all, it seems
at least approximately correct to say that S’s truth is a theorem of the semantic rules of
language L just in case it is analytic that, given the semantic rules of L, S is true.
But here we must be careful. Remember that there are two kinds of analyticity:
there is formal truth and there is also conceptualo truth. Given this, suppose that s1...sn
are the semantic rules of L. Even if S is formally true in L, and is therefore analytic in
that sense, the following sentence is not formally true in L:
(S1) “given s1...sn, S must be true.”
S1makes a true statement about the semantic rules of L. For many reasons, it follows
that S1 cannot itself be assigned truth by those semantic rules. (Some of these reasons
are strictly philosophical in nature, and others have to do with well-known results in
formal semantics.) At the same time, S1 obviously isn’t empirical or a a posteriori. (S1 is
obviously analytic one of the two meanings of that word.) Thus, S1 is either a formal
truth of some language L1 that is distinct from L, or S1 is a non-formal, conceptualo
truth. Let us consider each case. If S1 is not a formal truth of any language, then S1 is
not analytic in the same sense as S. Whereas S is analytic in the sense of being for-
mally true, S1 is analytic in the entirely distinct sense of being conceptuallyo true. And,
of course, there isn’t anything circular about defining “analytic” (formally true) in
terms of “analytic” (conceptuallyo true).
What about if S1 is a formal truth of some language L1? In that case, everything
that we just said about S1 is true of the sentence:
(S2) “given s*1...s*m, S1 must be true”,
where s*1...s*m are the semantic rules of L1. And, of course, what we just said about S
and S1 might also hold of comparable sentences S2, S3, and so on. But this process can-
not go on forever. Given any one language Li, it can be coherently supposed that its
expressions are given meaning by the expressions of some other language Li+i. But it
cannot be coherently supposed that this is true of every language. Linguistic meaning
ultimately has non-linguistic sources. So it is incoherent to suppose that a formally
true sentence, e.g. “either is is snowing or it isn’t”, is given meaning by expressions be-
longing to a language whose expressions are given meaning by those of yet another
 Conceptual Atomism and the Computational Theory of Mind

language - and so on ad infinitum. This means that the sequence of sentences described
earlier (S, S1, S2...) must come to an end, which in turn means that, for some finite n, Sn
is analytic (on one disambiguation of that word) but not formally true.
Formal truth presupposes conceptualo truth. The concepto of a semantic theorem
is to be understood in terms of conceptualo truth. The defender of Quine’s argument is
right to hold that the concepto of a semantic theorem is to be understood in terms of
the concepto of analytic truth. But we are not on that account guilty of circularity in
analyzing the concepto of analytic truth in terms of the concepto of a semantic theorem.
For there is nothing viciously circular about analyzing analytic (formal) truth in terms
of analytic (conceptualo) truth.
Let us sum up before we move on. The concepto of analytic (formal) truth is to be
understood in terms of the concepto of a semantic theorem. The concepto of a semantic
theorem is to be understood in terms of the concepto of analytic (conceptualo) truth.
Obviously there is no circularity in analyzing the concepto of analytic (formal) truth in
terms of the concepto of analytic (conceptualo) truth.
But now the question arises whether the concepto of analytic (conceptualo) truth
must be understood in terms of itself. If so, then the concepto of analytic (conceptualo)
truth is viciously circular. And in that case, the concepto of analytic (formal) truth is
also viciously circular, since that concepto, as we just saw, is to be analyzed in terms of
the concepto of conceptualo truth.
But the concepto of conceptualo truth is not to be analyzed in terms of itself, and is
not otherwise viciously circular. The sentence “there is no law where there is no gov-
ernment” is analytic because it is conceptuallyo true. And it is conceptuallyo true be-
cause, though true, it doesn’t register any empirical facts.
Thus, a conceptuallyo true proposition is one that is non-empirically true; and the
concepto of conceptualo truth need not, and ought not, be understood in terms of the
concepto of synonymy or of analyticity.
Quine might, and probably would, deny that there are non-empirical truths. But,
in this context, such a denial would be question-begging. After all, the very thing that
Quine is trying to show, in his attempt to eliminate the analytic-synthetic distinction,
is that there are non-empirical truths. A sentence is analytically true just in case it is
non-empirically true. Analytic truth is non-empirical truth. According to nominalism,
all facts are spatiotemporal and there is thus no non-empirical truth. So in assuming
the truth of nominalism, Quine is simply assuming that there is no analytic truth and
is thus question-beggingly assuming the very thing that he wishes to establish.
Let us sum up. The thesis that there are no analytic truths is ambiguous, as it can
mean either (a) that there are no formal truths or (b) that there are no conceptualo
truths. Given either disambiguation, Quine’s argument fails. Conceptualo truths are
non-empirical truths. So the concepto of “analytic” (conceptualo) truth isn’t to be under-
stood in terms of synonymy or, more importantly, in terms of itself; and that concepto
is therefore free of any vicious circularity. The concepto of analytic (formal) truth is to
be understood in terms of the concepto of a semantic theorem, and that concepto, in its
 Chapter 16.  Fodor’s first argument for conceptual atomism 

turn, is to be understood in terms of the concepto of conceptualo truth. Given that, as we

just saw, the latter concepto is free of vicious circularity, it follows that the same goes for
the concepto of formal truth. Thus, Quine’s thesis fails on each interpretation of it.
A brief look at philosophical history may shed some light on this topic. Leibniz
held that all the laws of logic can be reduced to the law of identity (now called “Leib-
niz’s Law”): if x and y are identical, then x has a given property iff y has that same
property. So, in Leibniz’s view, all logical truths – e.g. if p, and if p then q, then q – are
really instances of Leibniz’s Law.
But Arthur Pap (1958) showed that in order to reduce truths of logic to instances
of Leibniz’s Law, one must presuppose the truth of logical principles other than Leib-
niz’s Law and, consequently, that principles other than that law are ineliminable com-
ponents of logic. Leibniz’s thesis is therefore wrong.
In holding that all logical truths can be modeled on “all bachelor’s are married”,
Quine is guilty of a fallacy similar to Leibniz’s. (Chomsky (1988: 525) makes a similar
point.) Not all logical (or analytic) truths express identities (“the class of bachelors is
identical with the class of unmarried adult males”) or partial identities (“the class of
bachelors is a subset of the class of unmarried entities”). There is no non-contrived
sense in which “legal obligation presupposes governmental protection” holds in virtue
of some synonymy-relation between two expressions; and any equivalence between that
sentence and an identity (or partial identity) could be shown only on the basis of ana-
lytic truths that are not identities. So, contrary to what Quine assumes, it is not the case
that, if there are analytic truths, they can all be modeled on “all bachelors are unmar-
ried” or on any other sentence whose analyticity derives from a relation of synonymy.
It is not hard to find independent corroboration for this criticism of Quine’s argu-
ment. “All bachelors are unmarried” is trivial. The same is true of any sentence whose
analytic status depends entirely on some instance of synonymy: “there are three feet in
a yard”, “women are adult female humans”, “banks are financial institutions.”
But many analytic sentences are not trivial, and non-trivial analyticities never
hold merely in virtue of synonymy relations. Consider the following sentences: “there
are continuous functions that cannot be differentiated at any point”, “laws are narrow-
scope governmental assurances of protections of moral rights”6, “a symbol-token is an
instance of a relation between morphology and socio-psychological context”, “the
minimum number of sides that a three-dimensional closed figure can have is five”,
“there is no complete and consistent formalization of arithmetic.” The analytic status of
any given one of these sentences cannot be understood in terms of the fact that one
term is synonymous with another. Unlike “all bachelors are unmarried”, these sen-
tences do not merely register synonymies, and that is why they are all non-trivial.
There are trivial analyticities that do not merely register synonymies, e.g. “green is
a color.” (Attempts were made to show that “green” could be defined in terms of “color”;

6. This was the thesis of my doctoral dissertation.

 Conceptual Atomism and the Computational Theory of Mind

but these attempts failed, and it is clear that they must fail.7) But all non-trivial analy-
ticities do more than register synonymies.
Quine’s mistaken thesis that analyticity is to be understood in terms of synonymy
can be understood as a distortion of the correct point that a sentence is analytic, on one
disambiguation of that term, iff it is semantically true. Synonymy-based analyticities
are among the more obvious cases semantic truths: “there are three feet in a yard” is
analytically true because a “yard” is defined as a distance of three feet. But they are not
the only examples of analytic truth.

Additional problems with Quine’s argument

We said that, on one disambiguation of the term “analytic”, a sentence is analytic iff it
expresses a logically true proposition. Quine did not believe in propositions or in
meanings generally, and would thus reject this characterization of analyticity.
At least part of the reason that Quine rejects the notion of meaning (and, thus, of
propositionality) is that he is a nominalist: he believes in spatiotemporal individuals, but
not in platonic abstracta – there are instances of properties, but no properties (except in
so far as properties can be identified with, or otherwise reduced to, instances of proper-
ties). In Quine’s view, there are noises and ink-marks, but no propositions (except in so
far as propositions can be constructed out of noises, ink-marks, and the like).
My own view is that nominalism is wrong. (We’ve already seen that, in assuming
the truth of nominalism, Quine is begging the question against those who believe in
analytic truth. But we are going to set this aside here, so as to develop a new criticism
of Quine’s view.) But we don’t have to refute nominalism to show that Quine’s position
is indefensible.
Consider the following definition of analyticity:
(*) a sentence (or sentence-token) S, belonging to language L, is analytic exactly
if, given the semantic rules of L, and given no other empirical information, S’s
truth follows.
Notice that (*) doesn’t involve the notion of a proposition, except in so far as the con-
ceptso semantic rule and following from involve that notion. But it would be absurd to
deny that there are semantic rules, even though there are different reasonable views as
to what semantic rules are. And it would be absurd to deny that statements ever follow
from other statements. Somebody who denies that statement is thereby affirming the
statement: the statement that nothing follows from S follows from the statement that S is
a statement. Since there is analytic truth if anything follows from anything, it is there-
fore self-defeating to deny its existence, and somebody who denies that statements

7. Wittgenstein (1975) deserves credit for making this point. In Kuczynski (2006) I discuss why,
in my judgment, Wittgenstein’s insight warrants a non-atomistic conception of conception
 Chapter 16.  Fodor’s first argument for conceptual atomism 

ever follow from statements is affirming the negation of his own view. Thus, contrary
to what Quine says, it is possible to give a non-circular analysis of the concepto of ana-
lyticity, and it is incoherent to deny that some truths fall into that category. Notice, by
the way, that (*) covers both conceptualo and formal truth, and is thus a non-disjunc-
tive, completely general analysis of analytic truth.
Most importantly, it is clear at an intuitive level that some truths are analytic. To
know that there is pain only where there is sentience, one doesn’t need empirical
knowledge, over and above such as is required to grasp that proposition (or, if you
don’t believe in propositions, to understand the corresponding sentence).8 Quine’s ar-
gument is too wobbly to demand that we repudiate this intuition.
Quine would say this intuition is deceiving us, just as Kant’s intuitions were de-
ceiving him when they led him to repudiate the very idea of a non-Euclidean space, or
a non-Newtonian mechanics, or a non-Aristotelian logic. But as Laurence Bonjour
(1998: 5–7) and Michael Dummett (1973: 592–595) show, it is not hard to find demon-
strable support for our intuitions. Bonjour and Dummett provide the same argument,
which I will now state.
For the sake of argument, suppose that no statements are analytic. (The term
“statement” may be taken to mean either sentence or sentence-token or proposition.) In
that case, no statement of the form S1 confirms S2 to degree n is analytic: every such
statement is synthetic, and can be established only on the basis of experience. By the
same token, for any statement S3 whose truth is empirically established, it is up to fu-
ture empirical findings to determine what degree of probability S3 gives to S1 confirms
statement S2 to degree n. For an exactly similar reason, any future empirical finding S4
will, without the assistance of yet other empirical discoveries, be incapable of confirm-
ing the statement that S1 confirms statement S2 to degree n is confirmed to degree m by
statement S3. And so on ad infinitum.
Nothing can confirm anything unless at least some confirmation-relations are ana-
lytic. This doesn’t mean that this or that specific statement (e.g. “there is no government
where there is no law”, “there is no pain where there is no sentience”) is analytic. (Actu-
ally, it does show that the statement “given that some statements confirm other state-
ments, it is analytic that there are analytic truths” is analytic. But with a few exceptions of
this sort, this argument provides no special support for the view that this or that specific
statement is analytic.) But it does show that the class of analytic truths is non-empty.
This brings us to another argument against Quine’s position. If the statement:
(NAT) there are no analytic truths
is analytic, then Quine’s position is false. If (NAT) is synthetic, then empirical research
is needed to show that it is true. So there can be no demonstration or non-empirical
proof of (NAT) if that statement is true. Therefore that statement is non-demonstrable if

8. An argument similar to the one just given is found in Dummett (1978: Chapter 22).
 Conceptual Atomism and the Computational Theory of Mind

true. Quine’s argument is obviously not empirical. So if Quine is right to deny that there
is analytic truth, then we must regard his argument for that position as fallacious.

Hempel on analyticity

Carl Hempel (1952: 26–28, 1965: 113–116) has provided an argument against the an-
alytic-synthetic distinction that, although ultimately fallacious, is considerably more
compelling than Quine’s and that, unlike Quine’s, gives us real insight into the con-
cepto of analyticity.
Scientific conceptso are capable of partial (though almost never complete9) opera-
tional definitions. “x has length L” can be partially operationally defined as
(1) “x can be made to coincide with object y”,
(where y is some other object, e.g. the standard meter-rod in Paris) or as
(2) “it takes light n nanoseconds to travel from one end of x to the other.”
Of course, stipulative definitions are analytic. So supposing that O has length L, it is
analytic, given (1) and (2), that:
(3) it takes light n nanosecond to travel from one end of O to the other,
and it is also analytic that
(4) O can be made to coincide with y.
But (3) and (4) are not analytically (or logically) equivalent. Given that the one is true,
it does not analytically follow that the other is true. If (3) and (4) have the same truth-
value, that is entirely a matter of a posteriori fact.
Any one operational definition is purely analytic. But when one attempts to give
multiple (partial) operational definitions of a concepto, one has to make sure that those
definitions are empirically consistent with one another – otherwise the concept defined
is incoherent. For example, there is nothing incoherent in supposing that light takes m
(≠n) time to travel from one end of O to the other, but that O can be brought into coin-
cidence with y. For the sake of argument, suppose that to be true. In that case, given both
(1) and (2), the statement “O has length L” entails both that L can, and also cannot, be
brought into coincidence with y. To avoid saddling the term “length” with a broken and
incoherent concepto, we must make sure that our various operational definitions of that
term are mutually consistent: and that can be done only through empirical work.
Supposing that we (partially) define “length” in terms of (1) and (2), but we haven’t
yet established the empirical equivalence of those definitions, we are left with a situation
where a true statement (“O has length L”) analytically entails two other statements,

9. See Hempel (1965: 123-134).

 Chapter 16.  Fodor’s first argument for conceptual atomism 

namely (3) and (4), but where it remains an empirical question whether both of those
statements are compatible. But, if it is supposed that there is a sharp distinction between
analytic and synthetic statements, this is not a tenable situation. If S1 is true, and ana-
lytically entails S2 and S3, then it is ipso facto settled whether both of S2 and S3 are true.
The distinction between the analytic and the synthetic appears to have broken down;
and the statement “if O has length L, then it takes light n nanoseconds to travel from one
end of O to the other” isn’t non-arbitrarily categorized as either analytic or synthetic.
What was just said about “length” is true of each member of an extensive and im-
portant class of expressions. Many terms begin as expressions of lay-discourse and are
later appropriated by science. Having been thus appropriated, they are redefined, and
usually thereby rendered both more precise and more comprehensive in extension, in
light of scientific theories. Terms like “heat” and “temperature” have existed as long as
language itself. But, as they are used by the lay-person, those expressions seem to have
no determinate meanings outside the extraordinarily narrow corridors of day-to-day
human experience.
Physics has established that, for any number n that corresponds to a temperature
that can be directly experienced by human beings, a substance’s having certain mo-
lecular properties is necessary and sufficient for its having a temperature of n°. (This is
actually not quite accurate. But in this context that slight innaccuracy is irrelevant.) On
this basis, the term “temperature” can be identified with, and thus defined in terms of,
certain molecular properties. This greatly increases the range of things to which deter-
minate temperatures can be assigned. It also brings statements of the form Fx has tem-
perature n°G into systematic connection with powerful and well-organized bodies of
scientific theory, thereby increasing the predictive and explanatory import of such
statements. It also gives greater precision to such statements: previously they were de-
fined either in terms of instrumental operations whose outcomes were not a function
only of the relevant object’s temperature (whether a column of mercury rises by a cer-
tain amount in a thermometer is only partly determined by the temperature of the
substance in which the thermometer is immersed) or, even worse, were understood in
strictly phenomenological terms (in terms of what the object in question “felt like”).
The prescientific usage of the term “temperature” lacked the uniformity of usage
characteristic of rigorous discourse and, at least arguably, failed to determine a unique
concept. The process of assimilating that term, and other terms in its category (e.g.
“heat”, “freezing”), into science generated a situation exactly parallel to that described
a moment ago in connection with “length.” What we have said about “length” and
“heat” is true of “mass”, “acid”, and practically any expression denoting an explanato-
rily significant class of phenomena (or objects). Thus, given any one of these terms, an
argument exactly analogous to that given in connection with “length” points to the
existence of a breakdown of the distinction between analytic and synthetic truth.
 Conceptual Atomism and the Computational Theory of Mind

Evaluating Hempel’s argument

Let us start with a purely exegetical point. Hempel’s primary concern is not to discuss
the viability of the analytic-synthetic distinction, but only to describe scientific method-
ology. (His remark about the analytic-synthetic distinction is almost made in passing.)
From that viewpoint, Hempel’s analysis is unexceptionable. But we can accommodate
Hempel’s correct methodological point without rejecting the traditional view that all
truths are either analytic or synthetic (and without rejecting the exceedingly plausible
view that at least some truths are analytic) and without therefore embracing the subse-
quent revisions of classical logic that would be demanded by a rejection of that view.
First of all, we must distinguish reference-fixing from meaning-giving. An opera-
tional definition can be seen not as giving the meaning of “length” (or “temperature”
or “mass”…) but rather as fixing its referent. Suppose that Smith is the one person who
came to our party yesterday wearing a ridiculous looking plaid sportcoat; and suppose
that Smith is also the one person who gave money to the charity that Brown started. In
that case, the statements:
(5) “Smith is the one person who gave money to the charity that Brown started”
and
(6) “Smith is the one person who came to our party yesterday wearing a ridicu-
lous looking plaid sportcoat”
fix the referent of Smith. There is some x such that, given either (5) or (6), “Smith is
bald” means: x is bald. At the same time, (5) does, whereas (6) does not, entail that
Smith gave money to Brown’s charity; and (6) does, whereas (5) does not, entail that
that Smith wore a plaid sportcoat on a certain occasion. Each of (5) and (6) is a correct
definition of “Smith.” At the same time, given the statement “Smith is bald”, it follows
from (5), but not from (6), that a bald person gave money to Brown’s charity. So we
seem to have a situation where “Smith gave money to Brown’s charity” both does, and
does not, entail “a bald person gave money to Brown’s charity”, and we therefore have
a situation that precisely parallels that described in connection with “length.”
But obviously there is no breakdown of the analytic-synthetic distinction. The
content of (5) is perspicuously given by the statement:
(5*) Somebody x uniquely gave money to Brown’s charity and “Smith” names x,
and the content of (6) is perspicuously given by the statement:
(6*) Somebody x uniquely wore a ridiculous plaid sportcoat on a certain occasion,
and “Smith” names x.
What (5*) and (6*) make clear is that, where each of (5) and (6) is concerned, the defini-
tion of “Smith” actually being given is simply: “Smith” names x, for some x. The reason
that (5) does, whereas (6) does not, entail that somebody gave money to Brown’s charity
 Chapter 16.  Fodor’s first argument for conceptual atomism 

is that (5) is not just a definition: it consists of a synthetic statement (“somebody x

uniquely gave money to Brown’s charity”) conjoined (to put it loosely) with a definition
(“x is Smith”). The purely definitional component of (5) is given by a bare singular prop-
osition (x is named “Smith”); the remaining part of (5) is part of the pre-definition, as
opposed to the definition per se, of “Smith.” What we just said about (5) is true mutatis
mutandis of (6). So, ultimately, each of (5) and (6) defines “Smith” in the same way,
namely: “Smith” names x, for some x. But (5) and (6) convey different pre-definitional
information, and this creates the conundrum, or illusion thereof, just described.
For obvious reasons, expressions cannot be defined in a vacuum; definitions must
be made in terms of phenomena that are known to both speaker and auditor. But it
doesn’t follow that those phenomena always enter constitutively into the definition it-
self. In some cases, they are among the means used to give the definition, and are not
internal to the definition itself.
What we just said about “Smith” is true of “length”, “mass”, “heat”, “hardness”, and
other such terms. Each of (5) and (6) can be seen as fixing the referent, not giving the
meaning, of the term “length.” If this is right, then the content of (1) is given by the
statement:
(1*) There is some property P such that anything x (of the appropriate shape) has
P exactly if x can be brought into congruence with object y, and “length L”
refers to P.
And the content of (2) is given by the statement:
(2*) There is some property P such that anything x (of the appropriate shape) has
P exactly if it takes light n nanoseconds to travel from one end of x to the
other, and “length L” refers to P.
So where each of (1) and (2) is concerned, the strictly definitional component is given by
the bare singular proposition: the expression “length L” refers to P. It is true that (1) doesn’t
have the same analytic consequences as (2). But this doesn’t indicate that a statement can
simultaneously entail, and fail to entail, some other statement. It merely corresponds to
the banal fact that different synthetic statements have different consequences.
The distinction between reference-fixing and meaning-giving has far-reaching
implications as regards scientific methodology. One example of this concerns a doc-
trine called “operationalism.” According to this doctrine, a term T is scientifically
meaningful iff it is operationally definable, i.e. if for any object (or ordered n-tuple of
objects) x, there is some observable phenomenon E and some observable operation O
such that T describes x iff E results when O is applied to x.
Hempel (1965: 123–134) argued compellingly that operationalism is untenable.
But the term “operationalism” is ambiguous, depending on whether the term “define”
is taken in the reference-fixing or meaning-giving sense. If it is taken in the latter
sense, Hempel’s arguments against operationalism are entirely cogent. But those argu-
ments are not cogent if “define” is taken in the former sense. At the same, “operational-
 Conceptual Atomism and the Computational Theory of Mind

ism” is eviscerated if it is taken to be the view that an expression T denotes a scientifi-

cally respectable concept exactly if observable outcomes of observable procedures may
fix the referent of T. It is easily seen that, relative to that definition of “definition”, there
is no unique operational characterization of any phenomenon that has observable ef-
fects and that, consequently, operationalism is reduced to the triviality that, in some
cases, a theoretical (or otherwise unobservable) phenomenon may have an observable
effect. So Hempel’s arguments against “operationalism” are cogent, so long as that term
is taken in a way that does not eviscerate it (and this of course means that Hempel’s
arguments are cogent tout court).

The ambiguity of “ambiguity”

But even if we leave aside everything just said – even if we set aside the distinction
between meaning-giving and reference-fixing – we must still reject Hempel’s argu-
ment concerning the analytic-synthetic distinction. If “length L” is defined in two non-
equivalent ways, then Fx has length LG will be ambiguous between two non-equivalent
propositions. Obviously there will be statements that follow from the one disambigua-
tion that do not follow from the other. But this would no more warrant a revision of
our views concerning the analytic-synthetic distinction than is warranted by the fact
that, on exactly one of its two disambiguations, “John went to the bank” entails that
John went to a financial institution. Hempel’s argument thus embodies a failure to take
into account the pedestrian fact that some expressions are ambiguous.
The term “language” is ambiguous between “diachronic” and “synchronic” mean-
ings, to use Saussure’s (1966) terminology. The statement:
(7) “The English language came into existence after the year 1100 A.D.”
is true on one disambiguation, but false on another. If by “the English language” is meant
the set of semantic rules that, at this instant in time, characterize the language spoken in
Australia, England, and so on, then that statement is false. But that statement is true if by
“the English language” is meant a series of sets of semantic rules such that, first, the set of
rules operative in Australia, New Zealand, and so on, is a member of that series and such
that, second, there is a certain kind of spatiotemporal continuity between any two install-
ments in that sequence. So (7) is false on the “synchronic” meaning of “the English lan-
guage”, but true on the “diachronic” meaning of that expression.
Words change their meanings. (There is “semantic shift”, as linguists put it.) This
inevitably creates ambiguity since, during the period of semantic shift, some people
continue to use the term in the old way. (In fact, semantic shift usually, though not
necessarily, occurs in consequence of an absence of uniformity of usage.) But while the
ambiguities associated with words like “bank” and “dumb” are easily recognized as
such, and thus tend not to lead to confusion, the ambiguities created by semantic shift
are less likely to be recognized as such, and what are in fact purely terminological dif-
 Chapter 16.  Fodor’s first argument for conceptual atomism 

ferences come to be seen as having a substantive dimension. (We might say that “bank”
is an instance of “synchronic ambiguity”, since its ambiguity is not a reflection of change
in its usage, and that “temperature” is an instance of “diachronic” ambiguity, since its
ambiguity between molecular, phenomenological, and other concepts is a reflection of
language-change.) Synchronic ambiguities are (relatively) stable: “bank” has meant
both river’s edge and financial institution for hundreds of years. By contrast, diachron-
ic ambiguities are unstable, almost by definition, and they typically quickly give way to
uniformity of usage and thus to synchronic non-ambiguity.
For obvious psychological reasons, if a term is used non-uniformly on Monday
and uniformly on Tuesday, people will tend to regard as deviant those Monday-usages
of the term that don’t conform to its Tuesday-usage. Semantic shift thus tends to create
the illusion that, during the unstable interim-period, people were simply using the
expression in question wrongly.
A related point is that people tend to mistake continuous changes for identities.
So, because semantic shift usually happens continuously, there is often little recogni-
tion that it has occurred. This creates the illusion that the term has not changed mean-
ing, and the differences between pre- and post-shift usage are then seen as embodying
changes of a substantive, as opposed to purely verbal, nature.
Hempel’s argument overlooks the ambiguity of the word “ambiguity.” Scientific
terms are in a state of flux, since they are redefined in consequence of changes in sci-
entific theory. This creates ambiguity. But because, for the reasons just described, the
ambiguities in question are not comparable to those characteristic of paradigms like
“bank” and “dumb”, we don’t recognize them as such. As a result, what is actually an
innocent fact (e.g. that, at a certain point in time, “temperature” was ambiguous be-
tween optical and tactile concepts) appears to have far-reaching ramifications for logic
(“no sharp line between analytic and synthetic truth”).
We must also keep in mind that even if Hempel has shown that the analytic-syn-
thetic distinction breaks down where some expressions are concerned, nothing he has
said warrants the view that it breaks down universally or that it otherwise breaks down
in a manner that would vitiate our use of that distinction.
 chapter 17

Fodor’s second argument for

conceptual atomism

Here is Fodor’s argument:

Suppose that the concepto knowledge were conceptuallyo complex, i.e. that it con-
sisted of other conceptso. In that case, for obvious reasons, it would consist of the
conceptso belief, true, justified, and perhaps also the conceptso non-accidental and
reliable. Supposing that the concepto knowledge does consist of these other con-
ceptso, it follows straightforwardly that cognitive operations with the concepto
knowledge should demand more intelligence and more energy than cognitive op-
erations with conceptso like justification, belief, and truth. After all, thinking knowl-
edge would necessarily involve thinking justification, truth, and so on, but not vice
versa. So thinking knowledge would necessarily be more demanding, in terms of
intelligence and energy, than thinking any of these constituent conceptso.
But the empirical data doesn’t bear this out. Experiments, as well as non-ex-
perimental (e.g. introspective) empirical data, show that people process claims
about knowledge more quickly and effectively than they do claims about justifica-
tion, belief, and the like. Most four year olds make true claims involving the con-
cepto knowledge. (“I know that dogs bark”. “I know that I have to go school tomor-
row”). But, apart from the odd prodigy, no four year old ever talks about
justification.
What we just said about the concepto of knowledge is true of conceptso gener-
ally. Any four year old can grasp true propositions about money (“I have a dollar
in my pocket”) and about linguistic communication (“Daddy just told me to leave
his office”). But how many four year olds grasp the conceptso into which, if our
philosophical analyses are right, conceptso such as money and language decom-
pose? Every four year old grasps the concepto of number (“I have more toys than
you do”). But how many four year olds grasp the conceptso into which Frege ana-
lyzes that concepto? And in the rare cases where some four year old prodigy does
grasp one of these conceptso, will he operate with it as skillfully or expeditiously as
he operates with the concepto of which it is supposedly a component? Will he
operate as skillfully with the concepto set of equipollent sets as he does with the
concepto number? Will he find the concepto of an audience directed intention as
easy to manipulate as the concepto of a word? Would anyone? Surely not.1

1. Fodor (1998: Chapter 3).

 Conceptual Atomism and the Computational Theory of Mind

I would now like to present two counter-arguments. These two arguments are similar and
also overlap in content. One is to the effect that Fodor’s argument involves a failure to dis-
tinguish between “conceptual” and “non-conceptual” cognition. The other is to the effect
that Fodor’s argument involves a failure to distinguish between personal and subpersonal
cognition. It may be that the second argument is just a re-articulation of the first.

Our first counter-argument

You obviously grasp conceptso. You grasp the conceptso knowledge, justification, truth,
causation, and various others. But does all cognitive activity consist in the manipula-
tion of conceptso? Is all mental activity conceptual? Dogs and cats have minds, and
those minds process information. But do dogs and cats grasp conceptso? A dog can be
aware of an individual stick, an individual bone, and an individual person. But does a
dog grasp the conceptso stick, bone, and person? Many would answer by saying “no.” If
that is right, then not all cognitive activity involves concepto-manipulation and, in fact,
conceptual activity would be a very evolved and derivative form of cognitive activity.
Of course, what we’ve said so far hardly constitutes an irrefutable proof that there
is non-conceptual cognitive activity. But right now I would like to operate on the plau-
sible assumption that there is such activity. That said, we will momentarily encounter
some good reasons to accept that assumption; and in chapters 22–23, we will encoun-
ter what I believe to be conclusive evidence of its correctness.
What is the difference between grasping instances of a concepto and grasping the
concepto itself? What is the difference between grasping individual instances of (say)
redness and grasping the very concepto of redness?
Here is the answer that I would propose. To grasp a concepto is to grasp what dif-
ferent concrete situations have in common. To grasp the concepto redness (or squirrel
or rock…) is to have some understanding of how red (or squirrelly or rocky…) situa-
tions resemble each other. If this is right, then grasping conceptso is not a prerequisite
to grasping instances of conceptso. The opposite is the case: grasping situations is a
prerequisite to grasping conceptso.
Suppose that you see what is in fact an instance of redness. Even if you have a
concept of (say) redness, it isn’t as though your perception is the result of your exercis-
ing this concept. Innumerably many chromatically different possible situations corre-
spond to your concept of redness. Given only that you are thinking of some object as
being red, your mental activity leaves it quite indeterminate what the actual color of
that object is. This suggests that your sense-perceiving a situation cannot possibly con-
sist in your exercising the concepts that you have.2
What we just said about your concept of redness is true of practically any other
concept. (I say “practically” in acknowledgement of a class of exceptions that we will

2. A similar point is urged by Evans (1982).

 Chapter 17.  Fodor’s second argument for conceptual atomism 

deal with in a moment.) It seems to be of the essence of concepts that they apply to
many qualitatively different situations. This must be understood aright. Your concept
of redness doesn’t just apply to objects of different shapes and sizes and temperatures:
it also applies to objects that have different colors. Similarly, your concept of triangular-
ity doesn’t just apply to objects of different colors and sizes; it also applies to objects
that have different shapes. (It is obvious how these points are to be generalized.) By
contrast, a sense-perception is not color- or shape-indeterminate.3 This suggests that
one’s having a sense-perception is not the same thing as one’s exercising various con-
cepts. Given that sense-perceptions obviously have content, it seems to follow that
some content-bearing mental activity is non-conceptual.
There is a point that we must consider before we take the next step in our argu-
ment. As McDowell (1994) points out, you do have concepts that, if exercised, would
completely determine the color of the situation that you are considering. Suppose that
you are looking at some red object, and you think to yourself: I want a car whose color
is that exact shade of red. Let C be the concept corresponding to the italicized expres-
sion. C completely determines the color that you have in mind. And, of course, we
could imagine similar concepts that completely determine a certain size or shape or
temperature or degree of conductivity…Given this, it might seem that sense-percep-
tion does consist entirely in an exercise of concepts.
But this is not the case. Concepts like C presuppose the having of sense-percep-
tions (or, in any case, of mental imagery). Unless the occurrence of “that” in that exact
shade of red is associated with a sense-perception (or a mental image) of a specific red
situation, C won’t pick out any shade of red. One might counter-respond by saying that
the just mentioned perception is itself constructed out of concepts of the kind that
McDowell describes. But this would be both implausible and viciously regressive. So
while McDowell may be right to say that we can construct a concept that completely
determines a certain color (or size or temperature…), any such concept is parasitic on
the existence of non-conceptual, but nonetheless content-bearing, mentation.
This line of thought suggests that concepts are conceptualizations of non-concep-
tualo content. A hamster can see any instance of redness as well as any person. But
people do, whereas hamsters do not, grasp the concepto of redness. This is because

3. This is a point that Berkeley made long ago in his A Treatise Concerning the Principles of
Human Knowledge.
 Conceptual Atomism and the Computational Theory of Mind

people do, whereas hamsters do not, have an understanding of what it is that various
different situations have in common.4
Some conceptso are constructed out of others. For example, the concepto smart,
brown, politically liberal cow is constructed out of the conceptso smart, brown, cow, and
so on. But if the line of thought just put forth is correct, all conceptso are ultimately
constructed out of non-conceptualo contents, and one’s grasp of a concepto is a concep-
tualization of one’s grasp of non-conceptualo content. The immediate analytic struc-
tures of some conceptso (e.g. smart, brown, politically liberal cow) are to be understood
in terms of other conceptso. But, in the final analysis, the analytic structure of a con-
cepto is to be understood in terms of non-conceptualo contents.
In light of these points, it is easy to reconcile the presumption that the concepto
knowledge has analytic structure with the fact that the conceptso into which it supposedly
decomposes (e.g. justification, truth, reliable mechanism, belief) are much harder to ma-
nipulate. First of all, the analytic structure of the concepto knowledge is found not in other
conceptso, but in the non-conceptual contents out of which that concepto is constructed.
So Fodor is wrong to assume that, if the concepto of knowledge has analytic structure, it is
given by a biconditional that has conceptso on both sides of the main-connective. There-
fore Fodor’s argument implodes, given that it depends on that assumption.
Fodor is probably right to say that the concepto of knowledge is easier to manipu-
late than the conceptso belief, justification, non-deviant causal mechanism, and so on.
But that is easily explained in terms of the fact that those conceptso are of a higher-or-
der than the concepto of knowledge. The more general a concepto is, the more intel-
lectual wherewithal is needed to grasp it. Practically any three year old grasps the con-
cepto of a whole number. Practically any twelve year old grasps the concepto of a
fraction. But only a few sophisticates understand what it is that all and only whole
numbers, fractions, real numbers, imaginary numbers, quaternions, relation num-
bers… have in common. The concepto number is thus of a higher order of generality
than the concepto whole number.
What exactly is it for one concepto to be of a higher order of generality than an-
other? As we’ve seen, any concepto ultimately decomposes into non-conceptual con-
tents. But, as we’ve also seen, some conceptso immediately decompose into others. Of
course, a concepto may decompose into conceptso that in their turn decompose into
other conceptso that, in their turn, decompose into non-conceptual contents. There is
clearly no limit to how many levels of conceptualo decomposition may mediate be-

4. Incidentally, this discussion doesn’t presuppose a Lockean-empiricist conception of concep-

tion. (Fodor 1981 has done an admirably good job of showing why such a conception is untena-
ble.) It could well be that it is in virtue of our innate cognitive endowment that we human beings
are able to home in on the property that is had in common by all and only those situations that
instantiate the property of redness. It could be that experience has only a “triggering” role in our
grasping this concepto. But that is of no importance in this context. What is important is that,
however it is that we come to grasp that concepto, doing so involves our seeing what different
situations have in common.
 Chapter 17.  Fodor’s second argument for conceptual atomism 

tween a concepto and the non-conceptualo contents into which it decomposes. C is a of

a higher order than C* iff more levels of conceptualo decomposition mediate between
C and its non-conceptualo decomposition than mediate between C* and its non-con-
ceptualo decomposition. For one concepto to be more “abstract” than another is for it
to be of a higher-order than the other. The concepto of knowledge seems to be less ab-
stract than the concepto of justification. If this is right, it is no wonder that the former
is easier to grasp and to manipulate than the latter. The concepto of money seems to be
less abstract than the concepto of a stable political order. If this is right, then it is no
wonder that three year olds grasp the first concepto but not the second.
Let us sum up. The evidence that Fodor cites may show that one’s knowledge-con-
cept doesn’t decompose into a justification-concept. But this doesn’t show that the
concepto of knowledge has no analytic structure, and it doesn’t show that one’s concept
of that concepto doesn’t involve a grasp of that concepto’s analytic structure. This is
because it is a distinct possibility that a concepto’s analytic structure is to be found in
its non-conceptualo, as opposed to its conceptualo, decomposition. In Chapters 22–23,
we will see that this possibility is an actuality.
Also, leaving this last point aside, if our second criticism of Fodor’s argument is
cogent, the experimental data cited by Fodor is perfectly compatible with the supposi-
tion that one’s knowledge-concept decomposes into one’s justification-concept.
Let us end this section by drawing some consequences of the position just de-
fended. It has generally been thought that so-called “conceptual analyses” give the de-
compositions of conceptso: to analyze a concepto is to say of what other conceptso it
consists. 5 We will see in Chapter 24, when we are discussing Peacocke’s analysis of
conception, that this is not the case. In fact, we’ve already seen some reason to believe
this. And if what we’ve said so far is correct, an instance of what we typically refer to as
a “conceptual analysis” (e.g. a circle is a closed planar figure of uniform curvature) shows
that each of two different conceptso is a correct conceptualization of a given bundle of
non-conceptualo content. The concepto circle is a conceptualization of non-conceptual
content. So is the concepto closed planar figure of uniform curvature. And both are con-
ceptualizations of the same non-conceptualo content.
We must remember that there are often different, equally correct, analyses of a
given concepto. The statement:
(a) x is a triangle iff x is the area bounded by three lines such that any two of them
intersect, but such that not all three of them intersect
provides an adequate analysis of the concepto triangle; and so does the statement:
(b) a triangle is a three-sided, straight-edged, closed figure.

5. See Langford (1942). This conception of conceptualo analysis pervades the view advocated
in Fodor (1998), as I discuss in my article “Another argument against the thesis that there is a
language of thought” (Kuczynski 2004).
 Conceptual Atomism and the Computational Theory of Mind

It is demonstrable that the concepto triangle is capable of infinitely many, non-trivially

different analyses. This fact is inexplicable if we say that conceptualo analyses show
how conceptso decompose into other conceptso. It is not possible for the concepto tri-
angle to decompose into the concepto side, and also to fail to so decompose. But given
that both (a) and (b) are adequate analyses of the concepto triangle, this is what we
must say if we take the view that an analysis of a concepto shows how it decomposes
into other conceptso. One could resist this by denying that both (a) and (b) are genuine
analyses of the concepto triangle – by insisting that, for example, (a) constituted an
analysis whereas (b) did not. But such a position would be completely arbitrary, and it
would be inconsistent with the fact that what counts as an illuminating analysis is a
function in large part of the context – of what our starting assumptions and dialectical
objectives are.
We can avoid both self-contradiction and arbitrariness by saying that conceptualo
analyses show that there are two different, but equally adequate, ways of conceptual-
izing a given bundle of non-conceptualo information. In that way, we don’t have to take
the arbitrary view that either (a) or (b) is not an adequate analysis, and we don’t have
to take the self-contradictory view that the concepto triangle both does, and does not,
decompose into the concepto side. As we will see in Chapter 24, this approach also
enables us to explain how conceptualo analyses can be informative; and we will also see
how the same is not true of the view that an analysis of a concepto shows how it decom-
poses into other conceptso.
If these remarks are correct, it is clear what we must say about analytic condition-
als, for example:
(c) if x is a case of knowledge, then x is a case of justified true belief
Given that (c) is correct, it does not follow, pace Fodor (1998), that the concepto knowl-
edge decomposes into the conceptso belief, justified, and true. The concepto knowledge
is a conceptualization of a certain bundle of non-conceptualo content. The concepto
justified true belief is a conceptualization of that same data. But the concepto justified
true belief is not a conceptualization only of that data: it also conceptualizes data not
conceptualized by the concepto knowledge. That is why, even though (c) is true, the
corresponding biconditional (x is knowledge iff x is an instance of true justified belief) is
false. Much of Chapter 24 will be spent substantiating the analysis just put forth.
Fodor assumes that there is no conceptualo structure at all where it isn’t possible to
give correct, analytic biconditionals. Supposing that there is no analytically correct,
non-trivial biconditional of the form x is knowledge iff x falls under C, it follows, ac-
cording to Fodor, that the concepto knowledge has no analytic structure at all. In the
previous chapter we encountered one reason to hold that this position is mistaken, and
if the view that we have been urging is correct, we have just encountered another. The
concepto knowledge surely does have analytic structure. When a Gettierologist judges
that a certain situation constitutes a counterexample to a proposed analysis of that
 Chapter 17.  Fodor’s second argument for conceptual atomism 

concepto, it is on the basis of what an intuition tells him about the structure of that
concepto. Suppose, for example, that I propose the following analysis of that concepto:
(d) x is knowledge iff x is a true, justified belief on the part of somebody who is
wearing a blue shirt.
It is easy to think of situations where somebody wearing a blue shirt generates a justi-
fied true belief that is not knowledge. (The situations described by Gettier are easily
adapted to shows this.) But how do you know that a given situation constitutes such a
counterexample? Because your intuition tells you so; and your intuition obviously con-
stitutes some kind of insight into the structure of that concepto. (Such an intuition
doesn’t embody the outcome of empirical research: you don’t have to do empirical work
to know that a given situation is a counterexample to (d). So we are obviously dealing
with an insight into the structure of the concepto in question.) Thus, Fodor’s position
that the concepto of knowledge has no structure is implausible in the extreme.
At the same time, Fodor may be right to say that it is not possible to give an adequate
analysis of that concepto. But this is easily explained, given the position we are advocat-
ing. By a “conceptual analysis”, Fodor means a (non-trivial) statement of the form x is
knowledge iff x falls under concepto C. We’ve seen some reason to hold that the analytic
structure of a concepto is not to be found in its liaisons with other conceptso, and is
rather to be found in its relations to non-conceptualo content. If this is right, there is
obviously no reason a priori to expect that there holds some biconditional of the kind
just described.6 In any case, there would be no reason to expect this unless one had some

6. In light of the points made in this section, a case can be made that Fodor’s first argument
for atomism is question-begging. Fodor starts with the assumption that there is analytic struc-
ture only where there is the possibility of a biconditional of the kind just mentioned. This is
tantamount to the assumption that, for any concepto C, there is some set of conceptso C1…Cn
such that x falls under C iff x falls under C1…Cn is analytically true and non-trivial. Assuming
that those conceptso don’t have infinitely complex decompositions, we sooner or later arrive at a
set of conceptso that are not decomposable. And from this it follows that there is some set of
simple and unanalyzable conceptso out of which all other conceptso are composed.
It is hard to believe that there is a set of conceptso of the kind just described. It is also hard
to believe that any given concepto has an infinitely complex conceptual decomposition. Given
how implausible both of these views are, it seems reasonable to seek alternatives to the assump-
tions that led to them. In this case, such alternatives are not hard to find. We reject Fodor’s as-
sumption that conceptualo analyses give the decompositions of conceptso, and we say that such
an analysis identifies different ways of conceptualizing a single content. Given that there are
different but equally adequate analyses of many conceptso, this is a position we would wish to
take anyway. Leaving aside the dubious view that conceptso are either analytically unstructured
or decompose into analytically unstructured conceptso, it becomes unclear what reason there
could be for holding onto the view that analyzing a concepto consists in showing how it decom-
poses into other conceptso. Basically, Fodor’s conception of analysis has such counter-intuitive
consequences that one would hold onto it only if one had a pre-existing wish to validate those
very consequences. To this extent, Fodor’s first argument for atomism is question-begging.
 Conceptual Atomism and the Computational Theory of Mind

assurance that non-conceptualo information was categorically capable of being concep-

tualized – some assurance, in other words, that for any piece of non-conceptualo infor-
mation, there was an equivalent piece of conceptualo information. And, as we will see in
Chapters 22-24, there is reason to think that this is not the case.

Our second counterargument

Every cognitively normal person is able to make delicate judgments about other peo-
ple’s emotions, about art and music, and about the behavior of physical bodies. But
only the most erudite people can even begin to articulate the grounds for these judg-
ments. One could conceivably take the view that it is only when one can articulate the
grounds for one’s beliefs that those beliefs are principled. But such a view would be
radically implausible.7 It is more reasonable to say what we are naturally inclined to say,
namely: such judgments are at least sometimes principled, but the principles in ques-
tion are not always easily articulated and, although they are represented at some cogni-
tive level, they often fall outside the scope of introspective awareness.
Consider the following scenario. Jones listens to performances of thousands of dif-
ferent pieces, one third of which are unmistakably instances of rock ‘n’ roll, the other two
thirds of which are unmistakably not instances of that genre. Jones has no difficulty dis-
tinguishing the pieces that are instances of rock ‘n’ roll from those that are not.
When Jones classifies a piece as being rock ‘n’ roll, he does so on the basis of its
melodic, rhythmic, and harmonic properties. (We are setting aside the skeptical pos-
sibility that Jones is simply making lucky guesses over and over again.) In classifying a
given piece as rock ‘n’ roll, Jones is not making a conjecture; he is not making an induc-
tive leap into the future. Rather, he is conceptualizing the data that is already at his
disposal. Given that piece X has such and such musical properties, it follows conceptu-
allyo or analytically that it is an instance of rock ‘n’ roll. Of course, it is not analytic that
X has such and such properties, and it is not analytic that X is an instance of rock ‘n’
roll. But given that X has such and such properties, it follows conceptuallyo, and not
inductively or probabilistically, that it is rock ‘n’ roll. This shows that the concepto of
rock ‘n’ roll has analytic structure, and also that Jones’ classifications embody a knowl-
edge of that structure.
Before we proceed, we should note that the points just made furnish a refutation of
Fodor’s first argument for atomism. Would Jones be able to define the term “rock ‘n’ roll”?

7. Astonishingly, this position has been taken. It has often been said that what we know can-
not exceed what we can say. See Blackburn (1984: 140) for an illuminating discussion of this
problem.
 Chapter 17.  Fodor’s second argument for conceptual atomism 

No. It is unlikely that anyone is able to give an unexceptionable definition of that term.8
But it obviously doesn’t follow that the concepto of rock ‘n’ roll music lacks analytic struc-
ture or that Jones’ concept of that concepto doesn’t involve an acute understanding of
what that structure is. It is interesting and significant that the term “rock ‘n’ roll” is hard
to define. But what we are to conclude from that fact is obviously not that the concepto of
rock ‘n’ roll lacks analytic structure. Similar points hold in connection with the concepto
of knowledge. This shows that Fodor’s first argument involves a non-sequitur. In a mo-
ment, we will consider two hypotheses as to what that non-sequitur might be.
Let us return to our evaluation of Fodor’s second argument by developing the
scenario described a moment ago. Jones studies musicology, and comes to be able to
give a partial definition of the term “rock ‘n’ roll.” He learns that a rock ‘n’ roll song has
such and such rhythmic properties, and that it tends to comprise certain chord pro-
gressions and modulations (e.g. if it starts out in a given key, it will probably modulate
to the sub-dominant, and then to the dominant).
Jones may never be able to operate with the terms “dominant” or “12:3 rhythm” as
efficiently or skillfully as he is able to operate with the term “rock ‘n’ roll.” But, as we’ve
seen, it doesn’t follow that the concepto of rock ‘n’ roll music lacks analytic structure,and
it doesn’t follow that Jones’ grasp of that concepto consists in anything other than a
knowledge of that structure. Fodor’s argument clearly involves a non-sequitur.
In the previous section, we discussed one possibility as to what this non-sequitur
might be.9 But there is another possibility. In light of the researches of Chomsky, Marr,
and many other cognitive scientists, it is overwhelmingly likely that, along certain di-
mensions, what is subpersonally known vastly exceeds what is personally known. It is
also likely that, in the subpersonal realm, information is transmitted and processed ex-
ponentially faster and more adroitly than it is at the personal level. Further, there is rea-
son to believe that the information involved in subpersonal cognition is of a higher grade
than the information involved in cognitive processes occurring at the personal level.10
Given that at the personal level one can grasp the concepto of knowledge without
grasping the concept of justification, it doesn’t follow that one can grasp the first tout

8. Of course, some musical dictionaries contain a so-called “definition” of that term. But no-
thing is easier than to find (or, if not to find, then to compose) songs that satisfy the definition
but that are very obviously not instances rock ‘n’ roll. The same is true in my experience of any
supposed definition of any musical any form.
9. In case a brief recap is needed, here it is. The concepto of rock ‘n’ roll music has analytic
structure, and Jones’ apprehension of that concepto consists in his knowing what that structure
is. But the analytic structure of that concepto is to be understood not in terms of its relations to
other conceptso but in terms of its relations to non-conceptualo information. The reason that the
concepto of rock ‘n’ roll is easier to handle than the conceptso into which it supposedly decom-
poses – e.g. the conceptso sub-dominant, 12:3 rhythm, augmented fifth – is that the latter are of a
higher-order than the former. If the concepto of rock ‘n’ roll music is of order n, then these other
conceptso are at least of order n+1.
10. Churchland (1981) argues for this point very effectively.
 Conceptual Atomism and the Computational Theory of Mind

court without grasping the second. So given only that, at the personal level, Smith
grasps the concepto of knowledge but not that of justification, it doesn’t follow that he
doesn’t grasp the latter at the subpersonal level. If we said that everything that wasn’t
grasped at the personal level wasn’t grasped at all, we’d have to throw out cognitive sci-
ence in its entirety.
Further, given only that, at the personal level, Smith’s knowledge-thoughts are
more dexterous and energy efficient than his justification-thoughts, it doesn’t follow
that the concepto of knowledge isn’t to be understood in terms of that of justification.
Since any case of understanding has deep subpersonal roots, it could be that Smith
does understand the concepto of knowledge in terms of that of justification, but that
this understanding is to be found at the subpersonal realm of cognitive activity. Sup-
posing this to be the case, it would be irrelevant whether, at the personal level, Smith’s
justification-thoughts were less efficient than his knowledge-thoughts. The relevant
question would be whether Smith’s subpersonal justification-thoughts were less effi-
cient than his subpersonal knowledge-thoughts. Since the experimental and intro-
spective data cited by Fodor leave it entirely open how that question is to be resolved,
we may conclude that Fodor’s argument does not go through.
Permit me to elucidate this last point. The fact that, at the personal level, justifica-
tion-thoughts presuppose more cognitive wherewithal than knowledge-thoughts could
be a reflection of the structure of one’s subpersonal knowledge of these various con-
ceptso. It could be that, given this structure, one’s subpersonal knowledge of the one
concepto is more easily accessed by the personal realm than is one’s subpersonal knowl-
edge of the other. In general, once we concede that there is subpersonal cognitive activ-
ity, the empirical data that Fodor cites doesn’t force us to part with the idea that one
grasps the concepto knowledge in terms of the concepto justification. (Ironically, Fodor
himself is one of the most passionate and effective advocates of the view that there is
subpersonal cognition.)
People who cannot discourse about a concepto are not necessarily completely ob-
livious to it. People who couldn’t even begin to discourse about the concepto of justifi-
cation show astonishing subtlety in distinguishing cases of justified belief from cases of
unjustified belief, and thus clearly have a rich, intuitive understanding of that concepto.
The evidence that Fodor cites shows that people may be much more skillful in using
words like “knowledge” and “sentence” than in using words like “justification” and “as-
sertion.” This evidence thus suggests that a discursive, verbally represented under-
standing of the one concepto may be easier to acquire than a comparable understand-
ing of the other. But it is a commonplace of psychology that what is cognitively most
pervasive is often what is last to be given discursive representation. It takes a highly
trained and extraordinarily gifted person to paint what everybody sees or to put into
words what everybody knows.
Fodor may be right to say that, from the viewpoint of the personal level of cogni-
tion, the concepto of justification is quite recherché, while the concepto of knowledge is
quotidian. But that tells us very little as to how those conceptso are generally cogni-
 Chapter 17.  Fodor’s second argument for conceptual atomism 

tively represented. From the viewpoint of the personal realm of activity, the conceptso
that are cognitively most fundamental are the hardest to master and to articulate. If we
focus on what people say about those conceptso, it is easy to believe that they are com-
pletely oblivious to them. But an acute awareness of those concepts is revealed in their
judgments and general conduct.
Psychologists sometimes distinguish between “procedural” and “declarative”
knowledge.11 I know that Paris is the capital of France. This is “declarative” knowledge.
I know how to tie my shoes. This is “procedural” knowledge. My ability to speak Eng-
lish is also procedural knowledge, and so is a composer’s ability to write music. It
seems that the same is true of a person’s ability to determine whether a composition is
a fugue and of his ability and to read other people’s facial expressions. It is not out of
the question to suppose that all of the knowledge that we describe as “instinctive” or
“intuitive” is procedural knowledge.12
Subpersonal knowledge is typically expressed in terms of procedural as opposed
to declarative knowledge.13 It is hard to articulate procedural knowledge. It is astonish-
ingly hard to verbalize how to tie a double-knot. (It is easy to show somebody how to
do so. But telling them is another matter altogether.) In fact, I would guess that one
could easily produce counterexamples to any attempt to articulate what it is that all
and only instances of tying double-knots have in common. But it is obviously not an
option to say that the concepto of tying a double-knot has no internal structure, and it
is not an option to say that one’s grasp of that concepto doesn’t involve an understand-
ing of what that structure is.
What we just said about one’s ability to tie a double-knot is true of one’s ability to
use words. Few, if any, can say when exactly it is appropriate to use the words “the” and
“of ” and “and.” Further, whereas any English-speaker can use these words faultlessly,
most would have a devil of a time using the expressions in terms of which their ap-
propriate use is explicated – e.g. “c-command”, “NP-deletion”, “dominant node” – or
even understanding those expressions to begin with. But unless Chomsky and other
cognitive scientists are very wrong, it doesn’t follow that people are altogether ignorant
of the structures of the conceptso corresponding to theoretical terms like “c-command.”
What follows is that they have difficulty reproducing at the personal level the knowl-
edge that they already have at the subpersonal level.
Fodor himself is acutely aware of the distinction between declarative and proce-
dural knowledge, and also of the distinction between personal and subpersonal knowl-

11. See, for example, Lewicki et al. (1992).

12. That said, there seem to be at least some instances of innate (and thus “instinctive”) decla-
rative knowledge. The stock example is that infants know that breasts provide milk. Whether
this is a bona fide case of declarative knowledge is a question that lies outside the scope of the
present work.
13. See Lewicki et al. (1992).
 Conceptual Atomism and the Computational Theory of Mind

edge. But this particular argument of Fodor’s fails to take these distinctions into ac-
count, and for that reason is far from probative.
In this chapter, we’ve distinguished between:
(i) conceptual and non-conceptual knowledge,
(ii) personal and subpersonal knowledge, and
(iii) procedural and declarative knowledge.
We’ve already seen some reason to hold that there is overlap between the extensions of
the conceptso non-conceptual knowledge and subpersonal knowledge. But given only
what we’ve seen, it would be premature to say that these two conceptso have identical
extensions. It remains a possibility that at least some subpersonal knowledge is con-
ceptual in nature. Indeed, if Chomsky’s views are at least approximately correct, the
subpersonal realm is replete with manipulations of conceptso.
We’ve also seen that the concepto of procedural knowledge is closely related to the
concepto of subpersonal knowledge. But that connection is not one of co-extensiveness.
Much knowledge-how is personal knowledge. At the personal level, I know how to speak
English, how to tie my shoes, and how to ride a bike. If cognitive scientists are right, this
knowledge-how expresses subpersonal knowledge-that. For example, my ability to speak
English derives from a subpersonal understanding of conceptso like NP-deletion and c-
command.14 But given that at least some knowledge-how is personal, it follows that the
extensions of the conceptso subpersonal knowledge and procedural knowledge do not co-
incide. At the same time, it doesn’t follow that those extensions are completely disjoint;
for it remains a distinct possibility that some knowledge-how is subpersonal.
It is not clear exactly what the connections are between the conceptso subpersonal
knowledge, non-conceptual knowledge, and procedural knowledge. But what is clear is
that Fodor’s argument is vitiated by a failure to take these conceptso into account. This
is ironic since few thinkers, if any, have done as good a job as Fodor in defending theo-
ries that posit subpersonal cognition.15

14. Many philosophers, e.g. Hacker and Baker (1984), ridicule the idea that abilities – e.g. the
ability to speak English or ride a bicycle – involve subpersonal knowledge of conceptso like NP-
deletion. So far as this scorn is not based merely on the view that such views simply must be
false, given how at variance they are with common-sense, they are generally based on arguments
found in Wittgenstein and, to a lesser extent, in Searle. We will consider those arguments in the
next chapter.
It should be pointed out that Fodor (1975) does a superb job of diminishing the plausibility
of the common-sense-based rejection of subpersonal cognition by showing how the details of
theories that posit such cognition are extremely well-supported by experimental evidence.
Chomsky (1988) puts forth a number of compelling reasons why it must be supposed that there
is subpersonal cognition, also of why it must be supposed that what is knowledge-how at the
personal level is an expression of what is knowledge-that at the subpersonal level.
15. See Fodor (1975).
 chapter 18

Fodor’s third argument for

conceptual atomism

Here is Fodor’s third argument:

For the sake of argument, suppose that conceptual non-atomism is correct. In that
case, for any concept C that you have, you have C in virtue of having various
other concepts C1…Cn. So, for example, if you have a concept of triangularity, you
have it in virtue of having concepts of straightness, flatness, number, and so on. If
you have a concept of Socrates, you have that concept in virtue of having concepts
of space, time, causality, hemlock, and so on.
Supposing that non-atomism is correct, how exactly is your concept of trian-
gularity constituted by your concepts of straightness, flatness, and number? How
exactly is your concept of Socrates constituted by your concepts of hemlock, sar-
casm, and personhood?
The answer is that, if non-atomism is right, your believing that x is a triangle
consists in your believing that x is a closed, planar, three-sided, straight-edged
figure (or, at any rate, in your having some other comparable belief, e.g. that x is
the area bounded by three straight lines such that any two, but not all three, of
them intersect). Your believing that x is Socrates consists in your believing that x
is a unique dialectician to have died of hemlock poisoning (or in your having
some other comparable belief). In general, if non-atomism is correct, your beliefs
about a thing are constitutive of your concept of that thing.
This means that non-atomism has a number of obviously false consequences.
Suppose that you believe of Socrates that he was a talented wrestler in his youth.
In that case, if non-atomism is correct, your believing that x is Socrates consists in
your believing (inter alia) that x was a talented wrestler at some point. In that case,
anything that is not a talented wrestler ipso facto fails to satisfy the description
encoded in your concept of Socrates. So anything that was not a talented wrestler
fails to meet the preconditions that, given the structure of your concept, would
have to be met for you to believe it to be Socrates. It follows that you couldn’t pos-
sibly learn, or even come to believe, that Socrates was not a talented wrestler.
If non-atomism is right, what you believe about the things that your concepts
are concepts of determines what your concepts are concepts of. A corollary is that
your beliefs about a given thing are frozen: what you believe about a given object
cannot change. A related corollary is that two people cannot have different beliefs
about a given object. Yet another corollary is nobody can ever be wrong about
anything: if an object doesn’t satisfy the description associated with some concept
 Conceptual Atomism and the Computational Theory of Mind

of yours, then that concept is not a concept of that object. So nobody can ever have
a wrong belief about something of which they have a concept. Thus, nobody can
ever have a wrong belief.
Of course, what a person believes about a given object can change, and people
can have divergent beliefs about a given object, and it is possible to have erroneous
beliefs about a given object. Non-atomism is false since it is inconsistent with
these obvious truths.1

For the sake of argument, suppose that my having a concept of the number two con-
sists in my knowing the truth of the existence-claim:
(2) There is some x such that x is the unique successor of the number one.
Given that I have a concept of the number two, I can then go on to have all manner of
beliefs, both true and false, about that number. I can believe falsely that Smith has two
mansions (when he in fact has none); I can believe falsely that two is not a prime
number; and so on.
A corollary is that, given that you and I both have concepts of that number – in
other words, given that each of us knows the truth of some proposition like (2) – there
is no limit to how much we may disagree about it. I may believe that Smith has two
mansions, and you may disagree. I may believe that the number two is the class of all
pairs of objects; you may disagree.
To be sure, if non-atomism is right, then in order for both of us to have beliefs
about the number two, we must share some beliefs. For us to agree or disagree about
the number two, we must both accept (2) or, at any rate, some other comparable exist-
ence-claim. This puts limits on how much, from a non-atomist’s viewpoint, two people
can disagree about the number two. We cannot disagree as to whether it is the succes-
sor of one; nor can we disagree as to whether it is a number.2
But it does not follow from the non-atomist’s position that you and I have to be in
complete lock-step as to what we think concerning that number. Nor does it follow
that a person’s beliefs about that number cannot change. All that follows from the non-
atomist’s position is that certain core beliefs cannot be given up.
But this is not an unreasonable requirement. Indeed, its negation seems unreason-
able. Could you really believe of the number two that it wasn’t a number, or that it was
actually the successor of four (as opposed to one)? It seems that, if you believe of x that
it is not a number, you ipso facto don’t believe that it is the number two, and that if you
believe of x that it is greater than four, you ipso facto don’t believe it to be the immedi-
ate successor of one.

1. Fodor (1998: Chapter 2).

2. Actually, given the points made in Chapter 9, it isn’t clear whether even this much agree-
ment is necessary. What is necessary that, for some x such that x=2, you accept some sentence
whose truth-maker (not content) has the form x has phi and that I accept some sentence – not
necessarily the same as yours – satisfying the same condition.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

Fodor argues that, if one concedes even this much, one must also concede that
everything that a person believes about that number is constitutive of his concept of it,
and that a person cannot change his beliefs about it – he cannot come to believe that
Smith has two cars, and then later come to believe that Smith actually has only one car.
But, as we have just seen, that argument doesn’t survive scrutiny. One does need to
have some correct beliefs about x to have a concept of x. But once one has that concept,
there is no limit to how wrong one’s views about x may be.

Our response to Fodor (continued)

In some cases, one’s concept of a thing has a linguistic (or, more precisely, a meta-lin-
guistic) component. As we saw earlier, sometimes one’s concept of Socrates consists in
one’s knowing the truth of some claim like:
(3) There is some x such that x is uniquely somebody whom Smith and Jones are
now avidly discussing, and “Socrates” refers to x.
Under these circumstances, two people’s core-beliefs about Socrates may be very dif-
ferent. The content of Green’s concept of Socrates is given by (3). But Urdu-speaker
Aaronson may grasp Socrates by way of a very different existence-claim, for example
(4) There is some x such that x is uniquely somebody whom Professor Charles
was excoriating a moment ago, and “Sukrat” refers to x.
So it seems that, contrary to what I argued a moment ago, what two people believe
about a given thing can differ without limit – two people can both think about Socra-
tes, even though they don’t even agree that “Socrates” refers to him. (At least one of
those individuals may not know the English word for Socrates.)
The idea that this consideration warrants a rejection of any part of our analysis is yet
another expression of the failure to distinguish literal from cognitive meaning and of the
pernicious effects of modeling the one on the other. Suppose that my concept of Socrates
consists in my knowing (3). Under that circumstance, I cannot, indeed, come to believe
that Smith and Brown were not talking about Socrates: as Fodor says, in so far as (3) is
constitutive of my concept of Socrates, I cannot come to believe that Socrates doesn’t have
the characteristics referred to in that proposition. For exactly similar reasons, supposing
that your concept of Socrates consists in your knowing the proposition:
(5) There is somebody x such that x is the main protagonist in a book I am now
reading, called The Republic, and “Socrates” refers to x,
you cannot come to believe that Socrates was not the main protagonist of The Republic.
(We have already seen why this claim is in no way discrepant with Kripke’s (1972) se-
mantic insights.) As we’ve noted, supposing that non-atomism is correct, two people
 Conceptual Atomism and the Computational Theory of Mind

cannot agree or disagree about a given object unless there are at least certain existence-
propositions as to whose truth they are in agreement. So how his is this to be recon-
ciled with the points that we endorsed not two paragraphs ago?
This question is not hard to answer, given the points made earlier concerning con-
ception and literal meaning. We’ve already seen how I have a concept of Socrates in
virtue of knowing (3), and how you have a concept of Socrates in virtue of knowing (4).
Further, we’ve already seen how, for some x, FSocrates has phiG literally means x has phi,
notwithstanding that no one can grasp that bare proposition (except by grasping some
other proposition that describes it). At this point, we need to remember what we said
about what it is for somebody to mean proposition P by sentence-token S. Suppose I
sincerely and literally say “Socrates was wise.” In that case, what I mean by that sen-
tence-token is what the sentence-token itself literally means (viz. the bare, singular
proposition x was wise, where x is Socrates). To be sure, what “Socrates was wise” con-
veys to me is some existence-claim more or less like:
(6) There is some x such that Smith and Jones are discussing x and “Socrates” re-
fers to x, and “Socrates is wise” means: x is wise.3
But when I sincerely utter “Socrates was wise”, what I am affirming is confined to what
is to the right of the semantic-operator. We’ve discussed at length how people instinc-
tively keep non-linguistic (pre-semantic) information to the left of the relevant opera-
tors, keeping only the relevant linguistic (semantic) information to their right. Suppos-
ing that you are the person to whom I utter that sentence, you must find your way to
its literal meaning by way of your concept of Socrates. (We are supposing that you
speak English.) This concept, though different from mine, will lead you to the same
literal meaning; and when you affirm or deny “Socrates is wise”, what you are affirming
or denying is exactly what I affirmed. The miracle of language is precisely that it allows
people who have different concepts of a given thing to communicate about that thing.
The peculiarities of the individual’s concept of that thing become irrelevant.
Let us bring these points to bear on Fodor’s view. First, Fodor is right to say that, so
far as (3) constitutes my concept of Socrates, I cannot cease to believe that Socrates is
not an object x such that, on the occasion in question, Smith and Brown were talking
about x. But given that fact, there is no limit to what I may come to believe or disbelieve
about Socrates. I can come to believe that he studied in Egypt, and I can later come to
retract that belief. Further, what we just said about me is true (mutatis mutandis) of you.
Given that (5) constitutes your concept of Socrates, you cannot come to believe that

3. I say “more or less like”, and not “identical with”, in acknowledgement of the fact that, for
reasons we’ve discussed, the passage of time demands that the indexicals in the sentences ex-
pressing one’s concept change so as to reflect changes in one’s spatiotemporal relation to the
object in question. So, for example, if (6) expresses my concept of Socrates on Monday, then on
Tuesday, the sentence expressing that concept will contain the word “yesterday” in the place
where, on Monday, the corresponding sentence contained the word “now.” See Frege (1918) and
Kaplan (1989) for a similar point.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

Socrates was not the protagonist of The Republic; but given this fact, there is no limit to
what you can subsequently come to believe, or cease to believe, about Socrates. Finally,
because each of us knows what is meant by tokens of “Socrates was wise”, even though
we grasp that literal meaning in different ways (because of the differences in our respec-
tive concepts of Socrates), that does not affect our ability to mean the same thing by ut-
terances of that proposition. For some x, the proposition x was wise is what both you
and I affirm through sincere utterances of “Socrates was wise.” And so long as I know
some proposition like (3) and you know some proposition like (5) – in other words, so
long as each of us knows some proposition that enables that person to compute the lit-
eral meaning of tokens of “Socrates was wise” – there is no limit to how much you and
I may disagree about Socrates. So even though both of our respective concepts of Soc-
rates are given by descriptively rich existence-claims, and are therefore to be under-
stood in strictly non-atomist terms; and, further, even though those existence-claims
are very different; and, finally, even though you and I may therefore have little or no
agreement as to what properties Socrates had; our ability to communicate with each
other about Socrates is in no way threatened. So contrary to what Fodor argues, non-
atomism is in no way incompatible with either (i) the fact that one’s beliefs about a
given thing can change, or (ii) the fact that one can have erroneous beliefs about a given
thing, or (iii) the fact that two people can have different beliefs about a given thing.

Is non-atomism committed to inferential role semantics?

Fodor argues that conceptual non-atomism is identical, or equivalent, with a doctrine to

the effect that, if x has content C, x’s having that content supervenes on, or is identical
with, x’s having a certain inferential role. So x is Smith’s belief that O is a triangle exactly
if, in virtue of having x, Smith is disposed to make certain inferences, e.g. he is disposed
to infer O has three sides and O is flat and so on. Let (IRA) be our term for the doctrine
just described, since that doctrine gives an inferential-role account of mental content.4

4. IRA is an extremely popular doctrine. It is advocated by Sellars (1963), Peacocke (1992,

1996), and Brandom (1998), to name but a few of its exponents.
Brandom (1998: 94–97) argues that Frege held an inferential conception of propositional
content. In defense of this, Brandom (1998: 94) cites the following passage from Frege:
“In my formalized language [Begriffsschrift]…only that part of judgments which affects the pos-
sible inferences is taken into consideration. Whatever is needed for a correct [richtig] inference
is fully expressed, what is not needed is…not”
The idea seems to be that what does not affect possible inferences is truth-conditionally inert and
irrelevant from the view point of a symbolic system whose purpose is to capture cognitive content.
Brandom cites another passage from Frege (Brandom 1998: 95):
“There are two ways in which the content of two judgments may differ; it may, or it may not, be
the case that all inferences that can be drawn from the first judgment when combined with cer-
 Conceptual Atomism and the Computational Theory of Mind

According to Fodor, conceptual non-atomism is either identical with (IRA) or at

least presupposes the truth of that doctrine. Supposing that a concept of flatness, of
straightness, of the number three, and so on, are constitutive of one’s concept of trian-
gularity, that is because to believe that O is a triangle just is to believe that O is flat,
three-sided, and so on. So given non-atomism, it seems reasonable to say that brain-
state x is one’s belief that O is a triangle exactly if, in virtue of having x, one believes, or
is disposed to infer, that O is flat, three-sided, straight-edged, and so on. It thus seems
that non-atomism and IRA are identical or at least equivalent.
Supposing that this is correct, and also that Fodor’s argument against non-atom-
ism is cogent, it follows that IRA has the same problems as non-atomism, and must
therefore be rejected. We have already seen that the second of these suppositions is
false. But let us set that aside for now, and let us focus on the question: is non-atomism
identical with IRA? It is worth knowing the answer to this question even if, as we have
argued, Fodor’s arguments against non-atomism are misguided; for it might turn out
(and, in fact, it will turn out) that IRA has problems that warrant its rejection, and that
warrant the rejection of non-atomism if non-atomism presupposes IRA.
First of all, Fodor is right to reject IRA. It is obviously false to say that a brain-state
(or anything else) has a given content in virtue of its inferential role. Since there are infer-
ences only where there is already content, nothing can have content in virtue of its infer-
ential role. After all, to make an inference is to judge that, given the actual or supposed
truth of some content (some proposition or unit of non-propositional information) the
truth of some other content follows. Thus, so far as, in virtue of having x, I have a ten-
dency to infer that O is flat or three-sided, that is because x is already a belief that O is a
triangle (or, in any case, is some belief that warrants, or that I at least believe to warrant,
the belief that O is flat and three-sided). Since there can be inferences only where there is
already content, it is absurd to suppose that it is ever in virtue of its inferential role that a
given brain-state (or anything else) has a given content.
In any case, the version of non-atomism that is being advocated in this work is not
equivalent with IRA, and doesn’t presuppose the truth of IRA. (Henceforth, we will use the
term “non-atomism” to refer only to the specific kind of non-atomism that we are endors-
ing.) According to non-atomism, my concept of the number two coincides with a belief in
some existence-claim – some claim like (2). And supposing that B is the brain-state real-
izing that belief, my believing that n=2 doesn’t consist in its being the case that, in virtue of
having B, I am disposed to infer that n is greater than one. My believing that n=2 is identical

tain other ones can always also be drawn from the second when combined with the same other
judgments. The two propositions ‘The Greeks defeated the Persians at Plataea’ and ‘the Persians
were defeated by the Greeks at Plataea’ differ in the former way; even if a slight difference of
sense is discernible, the agreement of sense is preponderant. Now I call that part of the content
that is the same in both the conceptual content. Only this has significance for our symbolic
language [Begriffsschrift].”
 Chapter 18.  Fodor’s third argument for conceptual atomism 

with my believing that n is the successor of one: the latter is not inferred from the former,
and B’s being a belief that n=2 does not consist in its having any inferential role.
Of course, given that B is a belief that n=2 (i.e. that n=the successor of one), I am
disposed to make certain inferences: I am disposed to infer that n is even, that n is less
than three, and so on. But, for the reasons given a moment ago, it would be absurd to
say B’s being such a belief consisted in its having a certain inferential role, and it would
be false to equate non-atomism with such a view.

Transitional role semantics

Nonetheless, there is a doctrine that is similar to IRA that is not absurd (but that we will
argue to be false, on the grounds that it is a version of functionalism). Suppose that, in
virtue of having brain-state x, I have a disposition not to infer, but only to form the belief,
that O is flat, closed, three-sided, and so on. In that case, it might plausibly be said that x is,
or realizes, a belief that O is a triangle. The idea, appropriately generalized, would be this.
Let x be some brain-state of person P. There are certain possible beliefs B1…Bn such that x
has content C exactly if, in virtue of having x, P is disposed to form all, or at least some
chosen subset of, B1…Bn. According to this view, the content of a brain-state (or anything
else) supervenes not on its inferential, but on its transitional, role. This doctrine thus pro-
vides a transitional-role analysis of mental content. Let us therefore refer to it as TRA.
Christopher Peacocke advocates a version of TRA. In due course, we will discuss the
viability of his analysis and of TRA-analyses generally. But right now I would like to ad-
dress a narrower question. Does TRA have the defects that Fodor ascribes to non-atom-
ism? Supposing that TRA is correct, does it follow that I cannot change my beliefs about
Socrates or that two people cannot disagree as to whether Socrates was a good orator?
No. What we said about non-atomism (in the context of defending it against Fo-
dor’s third argument) is true mutatis mutandis of TRA. Let B be some brain-state of
mine; and suppose that, in virtue of having B, I am disposed to accept O is flat and O is
three-sided, and other such propositions. In that case, according to TRA, B is a belief
that O is a triangle. Now suppose that I come to believe (falsely) that O is green. In that
case, given a belief that x=O, I do indeed have a disposition to believe that x is green.
But does it follow that I am incapable of grasping the very idea of O’s not being green?
Does it follow that I mistakenly believe greenness to be definitive of O? No. Once again,
we must distinguish between the information needed to lock onto a concept and the
information that one ascribes to that concept in the aftermath of that lock-on. Suppos-
ing that TRA is right, given only that, in virtue of having B, I have a disposition to be-
lieve O is flat, O is three-sided, and so on, it follows that I grasp the proposition O is a
triangle. Those initial dispositions fix what B’s content is; they guarantee that B realizes
a belief in the proposition O is a triangle. Once I have that belief, I am at liberty to form
new beliefs, both true and false, as to what kind of a triangle O is and as to what bearing
O’s being a triangle has on other (supposed) facts. I can form the belief that O is a green
 Conceptual Atomism and the Computational Theory of Mind

triangle, and then later retract that belief (when I find out that I was wearing green
spectacles when I was looking at O). I can form the belief that O’s being a triangle neces-
sarily involves at least one of its angles having a measure of 90°, and I can later retract
that belief. I cannot, it is true, reject the idea that O’s being a triangle involves its being
three-sided or flat. But with those qualifications, and others of a similar nature, I can
form any belief at all as to what O’s being a triangle involves. So, by itself, TRA does not
have any of the unwelcome properties ascribed by Fodor to non-atomism or to IRA.
This is not to say that TRA is correct. In Chapter 24, we will find that TRA is false.
But given only what Fodor says, there is no reason to reject TRA. All forms of non-
atomism – whether or not they involve either inferential- or transitional-role accounts
of mental content – presuppose that there are some limits to what one can believe or
fail to believe about the objects of one’s concepts. But, contrary to what Fodor holds, it
doesn’t follow that, according to non-atomism, every belief that one has about a given
object becomes constitutive of one’s concept of it.

Why non-atomism is prima facie guilty of vicious regressiveness

There is a significant problem with non-atomism, but it isn’t one of the problems at-
tributed to it by Fodor. Let us start by considering our specific version of non-atom-
ism; we will subsequently show that what we say in connection with our analysis is true
of non-atomism generally.
According to our analysis, my concept of Socrates consists in my knowledge of
some existence-claim. But in order to grasp an existence-claim, one must already grasp
various conceptso. To grasp the proposition somebody was uniquely a great dialectician
to die of hemlock-poisoning, I must grasp the concepts hemlock, dialectician, and so on.
We seem to be launched on an infinite and vicious regress if we say of our grasp of
those conceptso what we said of our grasp of Socrates. To avoid that regress, we must
say that, ultimately, conception does not consist in knowledge of existence-claims. But
in that case, our analysis is erroneous.
All existing versions of non-atomism analyze the having of a given concept in
terms of the having of other concepts. For example, Peacocke (1992, 1996) says that
brain-state B is one’s concept of addition exactly if, in virtue of having B, one is dis-
posed to make certain judgments. Each of those judgments involves grasping some
proposition, and thus involves grasping the various concepts of which that proposition
is composed. So conception is analyzed in terms of conception, and we thus appear to
be stuck with a vicious regress (or circle).
This point is an extremely important one, and we will deal with it at length. Fur-
ther, we will find that it does invalidate certain kinds of non-atomism. We will also see
that, even though it does not invalidate our version of non-atomism, it does force us to
make substantive additions to that analysis. Drawing heavily on Evans’ (1982) distinc-
tion between “conceptual” and “non-conceptual” content, we will make those addi-
 Chapter 18.  Fodor’s third argument for conceptual atomism 

tions in Chapters 22 and 23. Here is what we will argue. There are different kinds of
concepts. There are concepts of properties; there are concepts of hyper-properties
(properties of properties); and there are concepts of individuals. There are important
differences between our concepts of individuals, on the one hand, and our concepts of
properties, on the other; and when we take these into account, the regress described a
moment ago vanishes.
In any case, it is crucial that we address the objection to our analysis just consid-
ered since we will otherwise remain oblivious to these important (and, we will see,
independently verified) stratifications in the concepto of a concept.

Terminological point

From now on, the word “concepto” (and its derivatives: “conceptso”, “conceptualo”, and
so on) will vanish from our lexicon. This is because we will almost always be discussing
concepts in the psychological sense only. In the rare instances when we do discuss the
latter, we will simply use the word “concept” (no sub-script), since the context will
make it clear what is meant.

Appendix to Chapter 18
The subpersonal and the subconscious

As previously noted, a number of philosophers reject the very concept of subpersonal men-
tation, believing it to be incoherent and not merely uninstantiated. The best and most in-
fluential arguments for this view are to due to Evans (1985: 322–344), Searle (1984, 1992),
and Wittgenstein (1958). Let us consider these arguments, starting with Evans’.
(EA) Thought is systematic. If you can think Bob is ambivalent towards Mary and
Seymour hates Ted, then you can think Mary hates Seymour and Bob is ambiva-
lent towards Ted. Put another way: if one grasps the concepts loves, ambivalence,
Ted, and so on, then one can grasp any permutation of those concepts. A person
can think any (viable) permutation of the concepts that he grasps.5 There is an
important corollary. Suppose that a person cannot think some proposition con-

5. Of course, it isn’t always clear what counts as a viable permutation of the concepts compo-
sing a given thought, and this makes it unclear whether our thinking is unlimitedly systematic.
Is the thought the property of being loving is ambivalent towards Bob’s ambivalence towards Mary
a viable permutation of Bob is ambivalent towards Mary and Mary loves Ted? If it is, then there
seem to be some limitations to how systematic human thought is, given that one can probably
grasp the second proposition without grasping the first. But, first of all, this objection may fail to
take into account the distinction between performance and competence. Second, and more im-
portantly, given Evans’ purposes, it is enough that thought be approximately systematic and that
 Conceptual Atomism and the Computational Theory of Mind

sisting of concepts C1…Cn. In that case, given that thought is systematic, it fol-
lows that one doesn’t grasp at least one of those concepts.
Of course, it is obviously an oversimplification to say that one can think any
combination of the concepts that one grasps. For example, no human being can
grasp the meaning of a sentence that is 27,000 words long, even though many a
human being grasps all of the concepts constitutive of the meanings of many
such sentences. But this doesn’t mean that human thought is only approximate-
ly systematic. We must remember Chomsky’s (1965) distinction between per-
formance and competence. Supposing that Smith is a highly intelligent person
who grasps concepts C1…C27,000, Smith’s inability to grasp some proposition
that comprises all of those concepts has nothing to do with his having a less
than perfectly systematic mind. It has instead to do with the fact that, for vari-
ous reasons, he is not able to fully exploit his own cognitive resources and, in
particular, his own ability to grasp any viable permutation of the concepts that
he grasps. Although Smith is intelligent, he is not infinitely so. Further, his
memory is imperfect, and he has only limited supplies of energy. It is because
he is subject to these limitations, and not because he has a less than perfectly
systematic mind, that Smith cannot grasp every (or even any) proposition con-
sisting of all of C1…C27,000. Smith’s “competence” – his cognitive ability – sur-
passes his “performance” – his ability to exploit his cognitive ability. In general,
given the distinction between competence and performance, there is no reason
to hold that human minds are less than perfectly systematic.
Now that we’ve taken care of these preliminaries, we can state our argu-
ment. If indeed there is subpersonal mentation, the concepts involved are cat-
egorically sealed off from the concepts available to consciousness and, more
generally, to the personal level of mentation. If cognitive scientists are right,
then our subpersonal minds are continually solving equations that would be
unsolvable at the level of personal ideation for anyone other than, and possibly
including, the world’s best mathematicians. If cognitive scientists are right,
there is a whole realm of concepts belonging to the subpersonal stratum that
are incapable of being deployed at the personal level. Of course, cognitive sci-
entists grant – nay, they insist – that one’s subpersonal grasp of these concepts
has effects on consciousness. It is because one can solve prohibitively difficult
differential equations that one has visual perceptions; it is because one can ap-
ply the most intricate sets of rules that one can learn, and then use, language.
But even though (if cognitive scientists are right) every moment of conscious
life embodies the effects of subpersonal mentation, and thus of one’s subper-
sonally grasping various (to the personal mind) abstruse and exotic concepts,
one’s subpersonal concepts are sealed off from one’s personal concepts.

we have an approximate understanding of what that involves. Delicate issues like the ones just
raised needn’t be resolved.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

In light of this, suppose that Smith can subpersonally, but not personally,
grasp the concept continuous function that cannot be differentiated at any point,
and that he can personally, but not subpersonally, grasp the concepts Fred and
completely baffling concept. In that case, Smith will not be able to grasp the prop-
osition: the concept of a continuous function that cannot be differentiated at any
point is completely baffling to Fred. Smith’s inability to grasp this proposition has
nothing to do with purely performance related issues. We are obviously dealing
with some kind of systemic block. But systematicity demands that Smith be
able to grasp just this proposition. Since the supposition that there is subper-
sonal mentation demands that systematicity be violated – indeed, that it be vio-
lated routinely – the very concept of such mentation is incoherent.
I believe that Evans is entirely right to hold that thought is systematic and that, leaving
aside performance-based problems, a person is (and must be) able to grasp any per-
mutation of the concepts that he grasps. But given only that thought is systematic, it
doesn’t follow that each of two separate cognitive units must be able to grasp any con-
catenation of concepts that belong either to the one unit or the other. So it doesn’t fol-
low that one must be able to combine a concept that one grasps at the personal level
with a concept that one grasps at the subpersonal level.
Suppose that Smith grasps the concepts Bob, Jane, and loves and that Jones grasps
the concepts Howard, Mary, and hates. Given only this information, the fact that
thought is systematic doesn’t demand that anyone or anything be able to grasp the
proposition Jane hates Howard or Mary loves Jane. What systematicity does demand is
that Smith be able to think Jane loves Bob and Bob loves Jane, and that Jones be able to
think Howard hates Mary and Mary hates Howard.
In the argument just presented, replace each occurrence of “Smith” with an occur-
rence of “Smith’s personal level of mentation” and replace each occurrence of “Jones”
with an occurrence of “Smith’s subpersonal level of mentation.” (If you wish, you may
also replace the expressions “Howard”, “Mary”, and “hates” with occurrences of expres-
sions denoting concepts that are more appropriately associated with subpersonal
thought: e.g. heavy NP-shift, c-command, differential equation.) The resulting argument
shows that systematicity does not demand that concepts grasped by a given person at
the subpersonal level necessarily be capable of being combined into units that can be
grasped by that same person at the personal level. Systematicity demands that con-
cepts within a given cognitive sphere or module be permutable. It doesn’t demand the
complete interpenetrability of every cognitive module.6
We should also point out that, if cognitive scientists are right, subpersonal concepts
combine and recombine with one another at least as freely as their personal counter-
parts. In fact, given how much information-processing is subpersonal, it seems clear that

6. My use of the term “module” is not necessarily meant to correspond precisely to what that
term has come to mean in cognitive psychology. So I am not using that term in the sense in
which it is used in, for example, Fodor (1983) – I am using it in a more generic sense.
 Conceptual Atomism and the Computational Theory of Mind

subpersonal concepts are much more promiscuous than their personal counterparts7,
and that the former are therefore less counterexamples to systematicity than the latter.
Evans takes it for granted that, were they to exist, subpersonal concepts would be
frozen – incapable of being used except in certain stereotyped ways. Given his correct
view that anything that was thus frozen would lack one of the essential properties of a
concept, he infers that there can be no subpersonal mentation. But Evans’ initial as-
sumption is wrong, given that subpersonal concepts, supposing that they exist, are
very much not frozen. Evans’ argument involves a failure to distinguish a concept’s be-
ing frozen from its being encapsulated, i.e. confined to a certain cognitive module or
sphere, and Evans’ argument implodes when this distinction is made.
We have seen that there is reason to doubt the cogency of Evans’ argument against
the view that there is subpersonal cognitive activity. But the thesis that forms Evans’
point of departure (namely, that concepts are not frozen) is a correct and deep one. In
the section following the next one, we will see that Evans’ point gives us another reason
to reject CTM. We will also find that (ironically) it helps us combat one of Searle’s at-
tacks on the thesis that there is subpersonal cognition.

Searle on the subpersonal: the homunculus problem

A more formidable challenge to the view that there is subpersonal mentation is given
by John Searle. According to Searle, the thesis that there is subpersonal mentation is
guilty of a version of “homunculus fallacy.” Here is a paraphrase of his argument.
Supposing that it exists, subpersonal mentation consists in the following of rules – in
the performing of derivations that accord with various logical and methodological
strictures. (In any case, the subpersonal mentation actually posited by cognitive sci-
entists like Chomsky and Marr consists in such acts of rule-following. Whether sub-
personal mentation, supposing it to exist, must consist in such acts is a question we
will set aside.) There can’t be rule-following without a rule-follower. So if there is
subpersonal thought, that means that there is a subpersonal rule-follower (to whom
we will refer as RF). But then we must explain how it is that RF can think. We gener-
ate a vicious regress if we say that RF’s cognitive achievements supervene on cogni-
tive performances that are buried in his subpersonal psyche. On the other hand, if
we say that we can explain RF’s ability to follow rules, and otherwise think, without
imputing subpersonal thought to him, we are conceding the explanatory uselessness
of the idea that there is such thought.8

Suppose that brain-state B realizes my thought that Socrates was wise, and that brain-
state B* realizes my later thought that somebody was wise. As we’ve discussed, for me

7. See Churchland (1981).

8. Searle (1992: 212-214).
 Chapter 18.  Fodor’s third argument for conceptual atomism 

to have inferred that somebody was wise from my belief that Socrates was wise, it is not
enough that B cause B*. It is necessary that B’s being a thought that Socrates was wise be
what is causally responsible for B*’s subsequent existence and, in particular, for its be-
ing a thought that somebody was wise. In fact, we’ve seen that even if this condition is
met, I still might not have inferred the one proposition from the other.
Nonetheless, there is, we saw, some special causal relation R such that (given that B
and B* realize, respectively, my thought that Socrates was wise and my later thought that
somebody was wise) I have inferred somebody was wise from Socrates was wise exactly if
B’s being a thought that Socrates was wise generates B* in manner R.
Given this, suppose that, in fact, B’s being a thought that Socrates was wise does
generate B* in manner R. In other words, suppose I have inferred somebody was wise
from Socrates was wise. And suppose that the inference is made at the personal level.
Where am I in all of this? Where is the person who makes the inference? If, by a person,
we mean some kind of an unpropertied entity underlying the causal transaction just
described, then that person is inert. By the same token, so far as that person is causally
involved, it is by way of his instantiating certain properties. (In this case, those proper-
ties would be those involved in having brain-states realizing certain kinds of beliefs)
So far as people are causally involved in anything, it is in virtue of their instantiat-
ing properties. But in that case, it is really property–instances, not people, that are
doing the causal work, except in so far as people are systems of interacting property-
instances.9 If some mysterious entity underlying the thoughts makes it be the case that

9. Of course, this view is due to Hume. It is now extremely unpopular in philosophical circles.
(See Merrick 2001: 122. Merrick points out that Strawson 1959: Chapter 3, Shoemaker 1997:
139, and Lowe 1989, 1996: 25 ff. are also opponents of the Humean view.) But, as we’ve noted, a
rejection of it seems to involve the incoherent idea of a self that lurks beneath various property-
instances. Even if selves are to be understood in this way, they don’t do any work, except in so far
as they instantiate the relevant properties. But in that case, as we just said, selves become com-
pletely irrelevant to understanding how the world works, since it is uncontroversially property-
instances that are doing all the work.
Also, we must distinguish the question “what is it for two thoughts to belong to the same
mind?” from the question “what is personal identity, i.e. under what circumstance are two events
or states of affairs constitutive of the same person?” The Humean view that selves (or persons) are
mere “bundles” of thoughts and perceptions is vulnerable to non-trivial criticisms. In general,
psychological conceptions of personal identity, such as Hume’s, are problematic and, at least ar-
guably (though not in my view), should be replaced by biological or organismic conceptions.
(See Hershenov 2002 and 2005.)
But none of this has any bearing on whether minds are anything over and above their
contents. The biological account of personal identity says that selves are not mere minds; it
doesn’t say that minds are mental contents plus biology. Given only that selves are more than
minds, and a fortiori are more than collections of mental contents, it would be absurd to say that
the same was true of minds. Thus, the view that minds might consist of mysterious selves under-
lying their various contents seems to me to derive from a failure to distinguish the question
“what are minds?” from the question “what are selves?”
 Conceptual Atomism and the Computational Theory of Mind

thought T (Socrates was wise) is followed by thought T* (somebody was wise), then so
far as that mysterious entity is doing any causal work, it is by way of its instantiating
certain mental properties. Past that point, that entity drops out. But that means that
what is doing the work are those property-instances. Positing an entity that “joins” the
thoughts is useless since, even if such an entity does exist, that joining can be accom-
plished only through the addition of more thoughts and, therefore, of more instances
of mental properties.
Also, positing such an entity is unnecessary. Given that B (or, more accurately, B’s
being a belief of a certain kind) brings about B* in manner R, it follows that an inference
has been made. We don’t need to bring in an unpropertied agent. We don’t need to adopt
the dubious view (found in discussions of free will) that there is “agent causation”, i.e. that
there is a mysterious spirit or unifying force beneath the thoughts that, by some miracle,
can synthesize the thoughts without itself instantiating any property and thus without
abiding by the general principle that it is instances of properties, and not bare particulars,
that are causally potent.
The inference just described occurred at the personal level. We saw that there was
an inference, but there wasn’t any inference-maker existing apart from the inference.
This doesn’t mean that people don’t make inferences. Obviously they do. But it means
that we cannot understand the making of inferences in terms of the activity of some
mysterious entity that exists apart from its own thoughts.
Suppose that B** is your belief that somebody was wise. For the sake of argument,
suppose that B precedes B**, where B is defined as before (namely, as my belief that Soc-
rates was wise). Obviously B’s being a thought that Socrates was wise cannot R-cause
B**’s occurrence. (In other words, B’s being such a thought cannot bring about B**’s oc-
currence in manner R.) And obviously that fact is constitutively linked with the fact that
B and B** realize thoughts had by different people. Instead of saying that B’s inability to
R-cause B** is a consequence of the fact that they mediate thoughts belonging to different
people, we may suppose that their belonging to different people consists in this fact. Sup-
pose arguendo that B’s being a belief that Socrates was wise could R-cause B**; suppose
that, because of some feat of neural surgery, our two brains were so linked that this were
a possibility. In that case, the boundaries between our minds would to that extent be
blurred. In general, for two thoughts or brain-states to belong to different people is for
them to be thus causally sequestered from each other, and for two thoughts or brain-
states to belong to the same person is for them not to be so sequestered.10
To put it succinctly: at the personal level, inference-making doesn’t involve an in-
ference-maker – at least not an inference-maker who is distinct from the act of infer-
ence-making in question. This is because inferences, like all mental events, are to some
degree constitutive of one’s mind and, therefore, of one’s self.

10. This is exactly the position that Russell (1940: 226-235) took. I am not claiming any origi-
nality here.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

This isn’t to say that non-mental (purely biological) entities are not thus constitu-
tive. Nor is it to say that statements like “John inferred that Bob had four cars” are
synonymous with, or otherwise analytically reducible to, statements about “bundles of
perceptions.” (Statements about tables are not analytically reducible to statements
about molecules, even though molecules are constitutive of tables.) It is merely to reg-
ister the fact that a person is not a kind of empty vessel which can be occupied by
thoughts, and that thoughts (and inferences, in particular) are ingredients of the caus-
al sequence on which a person’s existence supervenes.
Given that rule-following at the personal level doesn’t involve a rule-follower, it
would obviously be deeply arbitrary to suppose that rule-following at the subpersonal
level required a rule-follower. Searle’s argument thus turns out to involve a false view
as to what it is for a person to make an inference (and to follow a rule); it involves the
view that the making of inferences involves the intervention of some mysterious entity
underlying instances of properties and that, when this entity intervenes, it does so in
propria persona and not by way of its instantiating any properties.11

11. Of course, it is not the case that, given any two thoughts T and T* had by a single person,
the one R-causes the other. First of all, there are simultaneous thoughts, and simultaneous thou-
ghts obviously don’t generate one another; second, there are thoughts T1 and Tn that belong to
the same mind such that T1 doesn’t R-cause Tn, even though, for each i (1≤i≤n), Ti R-causes Ti+1;
and, finally, there are non-simultaneous thoughts T1 and Tn such that the condition just descri-
bed is not fulfilled. (What I am saying borrows heavily from Russell (1940: 226-235), and my use
of the term “R-connected” is obviously an echo of his term “R-sequence.”) Given any two thou-
ghts T1 and Tn, let us say that they are R-connected iff, for each i (1≤i≤n), Ti R-causes Ti+1. Given
this, I propose that two thoughts belong to the same person iff there is some thought Tm such
that both are R-connected to Tm (it doesn’t matter what direction the connecting goes in).
Here is an illustration. I think grass is green at time t. A moment later, I think I want milk.
There is no special connection between these two occurrences. But there would be no difficulty
finding a thought Tm such that both of these thoughts were R-connected to Tm. Tm might be a
memory that I had thought grass is green and then subsequently wanted milk. Here some might
make the following objection to our analysis:
“Can’t we conceive of two thoughts had by a single person being totally isolated from each
other? What about cases of split-personality? What about the distinction between subpersonal
and personal thought? Don’t these phenomena show that two co-personal thoughts can fail to
be R-connected.”
No. As we’ve discussed, different modules of a person interact. Of course, by the definition of
“module”, information doesn’t flow as freely between modules as it does within molecules. But
different modules do exchange information. (There would be no reason to posit their existence
if they didn’t.) The information in my visual module obviously interacts with the information in
my language module (I am supposing, for expository reasons, that there is a single visual mo-
dule and a single language module). If those modules didn’t interact, then I couldn’t read and I
couldn’t say anything, at least not anything worth saying (I couldn’t say, upon seeing a fire, “call
the fire-department!”). Further, the way in which information is exchanged between those two
modules is obviously much more direct than the way in which information is exchanged between
 Conceptual Atomism and the Computational Theory of Mind

two different people. The difference between intra- and inter-personal communication is that in
the former case the flow of information is mediated by relatively directly connected neural (and
cognitive) pathways, whereas in the latter case the transmission of information involves neu-
rally and cognitively dead intermediaries (inanimate sounds, ink-marks, and so forth). The dif-
ference, in other words, lies in the mode of causation. In the one case, but not the other, causa-
tion is relatively direct. The difference doesn’t have anything to do with the fact that, in the one
case, there is some one underlying entity that “has” both sets of thoughts, whereas in the other
case, the sets of thoughts are had by two such mysterious entities. It is true, of course, that one
person is involved in the one case, and that two people are involved in the other. But what this,
in its turn, consists in is a difference in the relevant causal mechanisms, and not in anything
having to do with some mysterious, unpropertied cognitive substrate.
What about cases of split personality? So far as such cases really do involve a complete se-
paration of two systems of thought from one another – so far as there really is no communica-
tion (of the relevant kind, viz. the kind that doesn’t involve cognitively dead intermediaries such
as noises and ink-marks) between Jekyll and Hyde – we are dealing with two people who happen
to share a body. And so far as there is communication (of the relevant kind) between the two
systems, it seems that the boundaries between their identities are fuzzy. It is, to my limited psy-
chiatric knowledge, an open empirical question whether two distinct people can inhabit a single
body. But it seems that this question deserves an affirmative answer precisely to the extent that
the same is true of the question “can there be two cognitive systems within a single body that are
as isolated with respect to each other as are the cognitive systems of two different people (e.g.
Kennedy and Johnson)?” So no matter what it turns out to involve, the phenomenon of split-
personality confirms our analysis.
Here is another, more substantial objection:
“You take it for granted that any two thoughts had by a single person are R-connected to some
one thought. But that seems like an empirical hypothesis that might turn out to be false.”
Let us start with a comparison. Simultaneous events cannot affect each other. But non-simulta-
neous events can stand in this relation – indeed, they must. More exactly, events E and E* are
non-simultaneous if there is a (nomically possible) causal sequence beginning with one of them
and ending with the other; and they are simultaneous otherwise. Of course, our awareness of
this principle was triggered by empirical developments in physics (in particular, by the asto-
nishing discovery that the speed of light is framework-invariant). But is it itself an empirical
proposition? Can we imagine two non-simultaneous but spatiotemporally related events E and
E* that are necessarily completely causally isolated from each other? It seems not. It seems that
being in the same “causal space” is constitutive of spatiotemporal relatedness. That may or may
not mean that non-simultaneous events must be connected by some actual causal sequence. (In
my view, it does mean this. Causal connections consist in disruptions of fields of energy. They do
not, contrary to what Hume seemed to believe, consist in sequences of discrete events. A field’s
intensity decreases continuously,never reaching zero. It follows that any two non-simultaneous
events are in fact causally connected. Reichenbach (1958) gave what I believe to be a completely
cogent argument to the effect that spatiotemporal order is to be defined in terms of causality. Of
course, if cogent, this argument shows only that given certain empirical facts – e.g. that all causal
transactions involve disruptions of energy-fields whose intensity doesn’t anywhere decrease to
zero – it is necessary that there be actual causal processes connecting any two non-simultaneous
 Chapter 18.  Fodor’s third argument for conceptual atomism 

Searle’s dispositionalism

Searle has another argument to the effect that behavior and conscious psychological
phenomena ought not to be explained in terms of subpersonal cognitive processes:
Let us start with a short story. Using a paper and pencil, you divide two multi-
digit numbers. (So you are doing “long-division.”) During the period of time that
you are carrying out this operation, the distance between Alpha Centauri and your
hand is increasing uniformly. Consequently, at any moment of time falling in that
period, the shape of your hand can be represented as a function of its distance from
Alpha Centauri. In other words, there is a rule that, for any instant in the time-
period just mentioned, assigns a shape to your hand on the basis of your hand’s
distance from Alpha Centauri at that instant. The motions of your hand therefore
accord with that rule, even though you are obviously not following it. The rules that
you are following have to do with facts about symbolism and mathematics that
have nothing to do with the distance of your hand from Alpha Centauri.
Human conduct – what noises people make and how they move their bodies
– may well be in accordance with the rules described by Chomsky, Marr, and
other cognitive scientists. But it doesn’t follow that anyone is following those rules.
So far as anyone thinks otherwise, he is failing to distinguish between following a
rule, on the one hand, and acting in accordance with a rule, on the other.

events. But, for reasons that I cannot go into here, I believe that it is necessary tout court that
non-simultaneous events be so connected.)
We must also remember Schlick’s (1919) point that, given a field-conception of causality, what
we would ordinarily describe as the location of an event is better described as the location of the
center of that event. Here, of course, the word “center” has both a geographic and a temporal mea-
ning. Given what we said a moment ago, there is no part of the universe that isn’t involved in a
disruption of a field. So a field-conception demands that every event be seen as being spread out
through all of space-time. It follows vacuously that any two events occupy the same region of
space-time. But the center of one event, i.e. the region of space-time where the disruption is grea-
test, may not – and, arguably, cannot – coincide with the center of some completely distinct event.
Similarly, if two thoughts are to belong to the same person, they must belong to the same “R-
causal space.” Maybe the objector is right; maybe two non-simultaneous co-personal thoughts
don’t have to be connected by an actual R-sequence. But given what we’ve said, it seems not un-
reasonable to say that they must be related to each other in a manner that is comparable to (and, in
fact, is a special case of) the manner in which two non-simultaneous events must be related.
There is another point to make. The objector presupposes that thoughts interact with one
another in much the same way that billiard balls (used to be erroneously thought to) interact. He
presupposes a “mechanical”, rather than a “field”, model of mental causation. If connectionist theo-
ries are correct, the field model is the right one, meaning that any mental event has “ripples” that
are felt in the furthest reaches of the subject’s mind. (See Cain 2002: 102-111 for a brief but effec-
tive discussion of the merits of connectionism, and also of some of the problems that it poses for
CTM.) This, of course, would corroborate my original view that any two non-simultaneous mental
events are connected by an actual R-sequence and are thus not merely in the same R-causal space.
 Conceptual Atomism and the Computational Theory of Mind

Of course, given only how people move their bodies, it doesn’t follow that they
are not following the rules described by Chomsky. But to explain human behavior,
there is no need to suppose that they are following those rules. So far as human
behavior is not to be explained in terms of what is occurring at the level of con-
sciousness, it can be entirely explained in terms of physiology. You ask me “how
did your interview go?” I respond by saying: “I thought that I answered some of
the interviewer’s questions creditably; but in other cases, my answers were spuri-
ous and the interviewer was visibly dissatisfied with them.” According to Chom-
sky, various unconscious cognitive processes mediated between auditory input
and verbal output. But there is no need for such an extravagant hypothesis, given
that a complete explanation of your response is obviously to be found in your
physiological structure, so far as it isn’t to be found in the contents of you con-
scious mind. There is simply no need to posit the sub-personal manipulations of
representations posited by Chomsky, and there is nothing to be gained by positing
them. There are conscious mental phenomena, and there is pure physiology; and
between the two, everything that needs to be explained can be explained.12

Searle is obviously right to distinguish between following a rule and acting in accord-
ance with a rule. But, beyond that, Searle’s argument has little force. By way of anticipa-
tion, the problem with Searle’s argument is that it involves a false view about confirma-
tion. More specifically, it falsely assumes that the hypothesis that so and so’s behavior
is guided by rule R is confirmationally interchangeable with the hypothesis that so and
so’s behavior merely accords with R. But we must cross many bridges before we can see
why exactly this assumption is false and how exactly Searle’s argument depends on it.
First of all, given only that there are two ways of explaining some phenomenon, it
doesn’t follow that one of those explanations is superfluous. Physics gives us one expla-
nation as to why Smith’s heart is racing. Biology gives us another explanation. But it
doesn’t follow that either explanation is wrong or even that either explanation is un-
necessary. Contrary to what Carnap (1934) argued, it is extremely unlikely that the
information embodied in biological explanations could be recovered by strictly physi-
cal explanations. As Jackson and Pettit (2004d) show, and as we discussed in Chapter
12, higher-level explanations capture principled regularities that lower level explana-
tions cannot capture, a consequence being that the former are richer in certain kinds
of information than the latter. For this reason, as Jackson and Pettit make clear, higher
level explanations are not merely inferior versions of lower-level explanations. Eco-
nomic explanations are not approximate psychological explanations. Physiological ex-
planations are not approximate microphysical explanations. And psychological expla-
nations are not approximate physiological explanations. So given only that there is a
strictly physiological explanation as to why I utter a certain noise, it doesn’t follow that
other explanations are impossible or unnecessary. In particular, it doesn’t follow that
Chomsky’s story is either wrong or unnecessary.

12. Searle (1992: 187-190, 227-254).

 Chapter 18.  Fodor’s third argument for conceptual atomism 

For the moment, let us consider a psychological explanation that does not involve
the supposition that there are unconscious (let alone subpersonal) psychological enti-
ties. You tell me that you want to drive to the store and you ask me where the car keys
are. I tell you that they are on my desk. You promptly walk in the direction of my desk.
Given what modern neuroscience tells us, there is every reason to believe that your
bodily movements can be understood in strictly physiological terms. At the same time,
your conduct can be understood, at least up to a point, in terms of what you desire (to
drive to the store) and what you believe (that the car keys are on my desk). Given only
that a physiological explanation is possible, it would be absurd to reject the psycho-
logical story. It would be equally absurd to reject Chomsky’s account of verbal behav-
ior, given only that a physiological explanation of that behavior is possible.
Of course, there are some obvious differences between the explanations offered by
folk-psychology and those offered by Chomsky. Consider my explanation of the fact
that you promptly start walking in the direction of the desk. I do not have any direct
awareness of the specific instances of belief and desire posited by that explanation. But
I have direct knowledge of cases where behavior relevantly like yours was caused by
mental states relevantly like those that I ascribe to you. After all, I know that your con-
duct is similar to my own conduct on past occasions and that, on those occasions, my
conduct was caused by mental states similar to those that I am now imputing to you.
Of course, my explanation may be wrong and even ill-founded. (It may be that you
have a history of feigning interest in driving to the store when, in fact, you have no
such interest.) But there is little doubt that some instances of that kind of explanation
are correct. By contrast, nobody has any direct knowledge of the processes posited by
Chomsky. In fact, nothing that anyone is directly aware of bears any significant resem-
blance to any one of those processes.
But given only these points, it doesn’t follow that such processes don’t exist. We are
not directly aware of anything similar to the phenomena posited by quantum physics.
But there is good reason to believe that such phenomena exist.
Searle has two objections to make to this last point. First, the concept of uncon-
scious mental activity is an incoherent one: there can no more be unconscious mental
phenomena than there can be square circles. Second, there is no empirical support for
the view that, when I speak, the sounds that I am making result from my subperson-
ally following various syntactic rules: in any case, there is as little empirical support for
it as there is for the view that the behavior of my hand when I write results from my
following some rule concerning Alpha Centauri’s state of motion.
We will deal with the first objection in the last section of this chapter. Right now,
let us focus on the second one. The problem with this objection is that it erroneously
assumes that the possible evidence in favor of the hypothesis that individual X is fol-
lowing rule R coincides with the possible evidence in favor of the hypothesis that X is
acting in accordance with R.
Before we can deal directly with Searle’s argument, we must make some general
points about the concept of confirmation. (These are not points that Searle disagrees
 Conceptual Atomism and the Computational Theory of Mind

with or overlooks. In fact, the points about to be made are platitudes that few would
disagree with. However, points relating to confirmation that Searle does overlook, and
that are relevant to his argument, can be understood in terms of these platitudes.) Sup-
pose we notice that Smith absolutely never walks on the grass. In light of this, consider
the following hypotheses:
(H1) Smith is deliberately following the rule stay off the grass. In other words, he is
(at some cognitive level) aware of that rule, and he is following it, and not just
acting in accordance with it.
(H2) Smith is not following the rule stay off the grass. He is following an entirely
different rule, namely: don’t walk on anything green.
(H3) Smith isn’t following any rule. Smith is an inanimate robot on whose forehead
there is a photosensitive cell that causes Smith to avoid walking on green surfaces.
No two of these hypotheses are confirmationally equivalent. In other words, it is not
the case that any evidence that confirms any one of these hypotheses equally confirms
the other. (In this context, “confirms” is to refer to either positive or negative confirma-
tion.) Ceteris paribus (H2) is more probable than (H1) if we find that Smith doesn’t
walk on green carpets. Ceteris paribus (H3) is more probable than either (H1) or (H2)
if we find that Smith never walks on green surfaces during the day, but that he does
walk on such surfaces at night – even if they are surfaces that any cognitively and per-
ceptually unimpaired human would know to be green (e.g. a golf course). It is only
within narrow limits that those three hypotheses are confirmationally equivalent.
Let us now consider a fourth hypothesis:
(H4) Smith is following the rule: either stay off the grass or become a square circle.
Despite what one might initially think, the possible evidence in favor of (H4) does not
coincide with that in favor of (H1). If Smith has very strange beliefs about mathematics
and about what people are capable of, he might walk on the grass even though he is
attempting to comply with the rule: either stay off the grass or become a square circle.
But ceteris paribus he will not walk on the grass if he is attempting to comply with the
rule: stay off the grass.
Setting aside concerns of a generally skeptical nature, it is often possible to estab-
lish beyond a reasonable doubt that another person has or lacks a given concept. (For
example, you can establish with reasonable certainty that a student of yours doesn’t
have the concept of a fractal.) Therefore, it could in principle be established that Smith
lacked the concept square circle, but that he had the concepts needed to grasp the rule
stay off the grass. Supposing that we learned this about Smith, his avoidance of green
surfaces would provide support for (H1) but not for (H4).
A few more preliminaries must be taken care of before we can deal head-on with
Searle’s argument. Let R be the rule earlier described that assigns a certain shape to my
 Chapter 18.  Fodor’s third argument for conceptual atomism 

hand depending on its distance from Alpha Centauri. For reasons similar to those just
given, the hypothesis that:
(H5) JM is following R
is not confirmationally equivalent with the hypothesis that:
(H6) JM is attempting to divide 174 by 13.
Given certain reasonable background assumptions, (H5) is strongly disconfirmed by my
not responding a certain way to questions like “name the five stars that are closest to the
Earth.” But given those same assumptions, (H6) is not disconfirmed by that same data.
Also, if I were trying to follow R, it is virtually impossible that I would succeed in
doing so. For various reasons, I couldn’t possibly know at each instant in the relevant
time-interval exactly how my hand was supposed to be moving. Second, even if I did
have that knowledge, my hand would still not move in quite the right way, given that I
do not have perfect hand-eye coordination.
Here are some simpler illustrations of the principle in question. My mis-adding two
twenty-digit numbers is consistent with the hypothesis that my behavior is guided by the
rules of arithmetic, but it is not consistent with the hypothesis that my behavior accords
with those rules. My never drawing a perfect circle is consistent with the hypothesis that
my behavior is often guided by the rule draw a perfect circle, but it is not consistent with
the hypothesis that my behavior accords with that rule.
When Chomsky says that we “follow” certain syntactic rules, he means that our
cognitive and physical activity is guided by such rules. That hypothesis is confirma-
tionally very different from the hypothesis that our conduct accords with such rules.
One reason why those hypotheses are different is that there is necessarily a gulf be-
tween “performance” and “competence.” Because we have limited energy, memory,
and intelligence, our attempts to follow rules don’t always succeed. So it is perfectly
consistent with Chomsky’s view that, on at least some occasions, our thought and con-
duct do not perfectly accord with the rules that, according to Chomsky, we are follow-
ing. It is precisely because being guided by a rule doesn’t involve acting in perfect ac-
cordance with it that Chomsky distinguishes between performance and competence.
We can now weigh in on the debate between Chomsky and Searle. Chomsky no-
ticed that our overt behavior and conscious judgments are consistent with the hypoth-
esis that linguistic activity results from our following certain kinds of rules. On this
basis, Chomsky hypothesized that our linguistic activity results from our following
rules of that kind. Searle grants that our linguistic activity may accord with the rules
that Chomsky describes. But, he argues, this gives us little or no reason to hold that we
are actually following those rules.
What Searle appears to overlook is that the evidence that supports the hypothesis that
we are following rules involving notions like c-command and heavy NP-shift is very differ-
 Conceptual Atomism and the Computational Theory of Mind

ent from the evidence that supports the hypothesis that we are simply acting in accordance
such rules. So supposing that R is a rule that involves such notions, the hypothesis that
(H7) Smith is following rule R
is not confirmationally equivalent with the hypothesis that:
(H8) Smith’s linguistic activity accords with rule R.
For example, (H7) is, whereas (H8) is not, consistent with the supposition that Smith’s
behavior doesn’t always accord with R. In fact, (H7) is consistent with the supposition
that Smith’s behavior never accords with R.
Let us illustrate these ideas by assigning a definite value to R. Supposing that S is a
sentence of the form F…the phi the psi… G, let R be the rule that says how S’s meaning
depends on the values of phi and psi. (Examples of sentences having that form are “he
saw the woman the old man warned him about” and “she threw paint on the car the
man fixed.” So R says how the meaning of the former depends on the meanings of
“woman” and “old man”, and it says how the meaning of the latter depends on the
meanings of “car”, “man”, and “fixed.”)
Smith’s being unable to assign the right meaning to the sentence “the girl the cat
the dog the boy fed bit saw fell” is inconsistent with (H8) but perfectly consistent with
(H7). In fact, if Smith is following R, it is a virtual certainty that his linguistic activity
will not always accord with that same rule, given the distinction between competence
and performance. So if Smith’s activity were to accord perfectly with R, that would be
evidence that he was not following that rule.
Conceivably, Searle might respond by saying the following:
Maybe you are right to say that X is following R is confirmationally different from
X is acting in accordance with R. And maybe you are right to say that, in some
cases, the empirical evidence is more consistent with the first hypothesis than it is
with the second. But it doesn’t follow that we have to embrace the dubious view
that we are subpersonally following rules. After all, the hypothesis X is following R
is confirmationally indistinguishable from X’s behavior is in rough accordance with
R. So why not just say that people are not subpersonally following rules but that,
for strictly physiological reasons, they can be described as though they are? It
seems to me that this is the most cautious and theoretically conservative path.
And it is demonstrably no less better supported by the empirical evidence than
Chomsky’s hypothesis that we are actually (albeit subpersonally) following vari-
ous recherché rules.

The problem is that, contrary to what the objector supposes, X is following R is in fact
confirmationally different from X is acting in rough accordance with R; and those confir-
mational differences favor Chomsky’s view.
Let R be some arithmetical algorithm. Ceteris paribus, I am more likely to follow
R correctly when I am adding small numbers than when adding big ones. Ceteris pari-
 Chapter 18.  Fodor’s third argument for conceptual atomism 

bus I am more likely to follow R correctly when I am well rested than when I am se-
verely sleep deprived. Ceteris paribus I am more likely to follow R correctly when the
relevant numerals are clearly visible than when they are hard to see.
The presence of a performance-inhibiting factor can typically be established inde-
pendently of the degree of success of the relevant performance. Your views as to whether I
am well rested probably don’t have to be based on the success of my attempts to add num-
bers. The same is true, of course, of your views as to whether I can see the relevant numer-
als and also of your views as to whether those numerals denote large or small numbers.
A corollary is that the absence of a performance-inhibiting factor can typically be
established independently of someone’s performance. Smith says that he is a world-
class sight-reader. But every time I ask him to sight-read an easy piece, his perform-
ance, while passable, is not that of a world class sight-reader. There are different pos-
sible explanations as to why Smith’s performance is mediocre. One is that he is not a
world-class sight-reader. Another is that, although he is a world-class sight-reader, he
was suffering from an acute vitamin B deficiency every time I asked him to play. Yet
another is that somebody paid him a million dollars to simulate the performance of a
mediocre sight-reader. If Smith really is suffering from a vitamin B deficiency, or
somebody has paid him off, that fact can, in principle, be established with reasonable
certainty. In any case, the hypothesis Smith is a world-class sight-reader but he was suf-
fering from a vitamin B deficiency every time I asked him to perform is confirmationally
very different from the hypothesis Smith is a mediocre sight-reader, and his mediocre
performance perfectly reflects his actual level of ability.
In light of these points, consider the following two hypotheses:
(a) X’s behavior is in approximate accordance with R. So far as X’s behavior accords
with R, it is because X is actually following that rule. So far as X’s behavior devi-
ates from R, it is because X’s performance is not up to his competence.
(b) X’s behavior is in approximate accordance with R, but X is not following R.
(a) and (b) are confirmationally very different. If X is actually following R, then his not
perfectly according with R is to be explained in terms of the presence of some per-
formance-inhibiting factor. Because the presence or absence of such a factor can be
established with reasonable certainty, the hypothesis that Smith’s performance is to be
explained in terms of his imperfectly following R is confirmationally different from the
hypothesis that his performance is not to be explained in terms of his following R. For
similar reasons, the hypothesis that our linguistic activity involves our following rules
involving concepts like verb-deletion and heavy NP-shift is confirmationally very dif-
ferent from the hypothesis that our approximate compliance with such rules is not to
be understood in terms of our following such rules.
Chomsky’s hypotheses are therefore not evidentially interchangeable with any hy-
potheses that don’t explain our behavior in terms of our following rules of the kind just
described. So the question whether Chomsky is right is an empirical one. (There is a
 Conceptual Atomism and the Computational Theory of Mind

qualification to this. Searle maintains that the concept of subconscious thought is sim-
ply an incoherent one. If he is right about this, then the question whether Chomsky is
right is not a strictly empirical one, and is at least partly conceptual. We will examine
this view, as well as Searle’s argument for it, in the last section of the present chapter.)
An important point made by Gareth Evans (1985: 322–344) helps make it clear that:
(H7) Smith is following rule R
is confirmationally very different from:
(H8) Smith’s linguistic activity accords with rule R,
and thus helps make it clear that Chomsky’s position is less frail than Searle makes it
out to be.
Evans points out that concepts and aptitudes are not frozen. If you speak Chinese there
isn’t going to be any one Chinese sentence that you understand. Of course, it is perfectly
possible to know the meaning of just one Chinese sentence. Somebody who doesn’t speak
Chinese can be told what some one Chinese sentence means, or he can figure it out on the
basis of context. But a genuine knowledge of Chinese will necessarily be “productive”, i.e. it
will be expressed in an ability to understand a multiplicity of different Chinese sentences.
What is true of the ability to speak a language is true of other abilities, e.g. the ability
to play the piano. You cannot learn to play a piece of any complexity – e.g. Mozart’s so-
nata in A-minor – unless you can play the piano. (Conceivably, a non-pianist could learn
to play “Happy Birthday” – but not a Mozart sonata.) And if you can play the Sonata in
A-minor, then (assuming that performance-inhibiting factors aren’t an issue) you can
learn to play other pieces, e.g. the sonata in C-minor. This isn’t to say that you will play
the sonata in C-minor as well as you can play the sonata in A-minor, and it certainly isn’t
to say that you can learn to play any piano piece that has ever been written. (Many pian-
ists who can play Mozart cannot play Liszt.) But it is to say that, if you try to learn the
sonata in C-minor, you won’t find that you hit a wall: you won’t find that you are com-
pletely incapable of playing the relevant sequence of notes.
What we just said about aptitudes is true of concepts.13 It isn’t as though somebody
who grasps the concept of addition can add three and five, but can’t add any other two
numbers, and it isn’t as though there will be only one way that a person can apply his
knowledge that 3+5=8. Suppose that Jones can figure out that Ted has a total of eight
shoes on the basis of the information that Ted has exactly three shoes in one closet and
exactly five shoes in another closet, and no shoes that are not in either of those two

13. In my view, aptitudes, e.g. the ability to play the piano, are expressions of subpersonal
conception. But this is a debatable point, and I won’t insist on it here. I provide some argument
for it at the end of Chapter 16. Chomsky’s work can be seen as an attempt to show that our
knowledge of language expresses subpersonal knowledge of various rules, and thus expresses a
subpersonal grasp of various concepts. (I am now in the process of trying to show that this at-
tempt on Chomsky’s part does not have the shortcoming that are sometimes ascribed to it.)
 Chapter 18.  Fodor’s third argument for conceptual atomism 

closets. In that case, Jones is going to be able to figure out that Larry has eight houses
on the basis of the information that he has exactly three houses in Jamaica and exactly
five houses in Brazil, and no houses that are not in Jamaica or Brazil.14
If somebody’s writing “12+15=27” results from his following the rules of arithmetic,
then he necessarily grasps the concepts involved in that rule, and his grasping those
concepts will be expressed in ways other than his writing that particular inscription. In
fact, it will be expressed in ways other than his writing any inscription. Supposing that
Smith grasps the concepts constitutive of R (where R is some rule, or body of rules, relat-
ing to arithmetic or semantics), he is going to make certain noises to his stock-broker
that he wouldn’t otherwise make; he is going to refrain from buying the luxury car that
he always wanted, and will instead buy an economy car; and so on. His grasp of those
concepts will not find expression only in his tokening sentences of arithmetic.15
If Smith’s writing “12+15=27” does not result from his following the rules of arith-
metic and semantics, he doesn’t necessarily grasp any of those concepts; and he won’t
necessarily engage in any of the behaviors just described. So (H7) and (H8) are sup-
ported by very different bodies of evidence. Of course, those two bodies of evidence
overlap. By itself, Smith’s writing “12+15=27” equally confirms both hypotheses. But if
he is really following the rules of arithmetic, and thus grasps the concepts twelve, addi-
tion, fifteen, and so on, then his behavior after writing that inscription will soon start to

14. In the movie Rain Man, Dustin Hoffman plays an idiot savant (named “Raymond”) who
can do extraordinarily difficult math problems, but is totally incapable of the most rudimentary
applications of that knowledge. If asked “what is 87,067 divided by 546?”, he can instantly come
up with the answer. But he draws a blank if asked “if Bob has one house, and Larry has two
houses, how many houses do they have altogether?” To my limited knowledge, that movie is
clinically accurate.
How do cases such as this one bear on Evans’ point that concepts are not frozen – that a
genuine understanding of addition or anything else is not going to be expressed in just one way?
Here are my thoughts on this. A grasp of the concepts of addition, multiplication, and so on, is
present at one stratum of Raymond’s mind. Further, that knowledge is not frozen with respect to
the other contents of that stratum of cognition. But even though it is unfrozen vis-à-vis the other
concepts in the stratum just mentioned, that knowledge is frozen vis-à-vis the contents of other
strata of cognition. In particular, that knowledge can interact only to a limited degree with the
contents of the personal stratum of cognition.
The mind is divided into different modules. (In any case, this is what Fodor 1983 asserts,
and I find his arguments compelling.) Those modules communicate. But the amount of commu-
nication between modules is miniscule compared to the amount of communication within a
module. So a piece of knowledge may be frozen vis-à-vis the bodies of knowledge that don’t
belong to the same module as that knowledge. But no piece of knowledge is frozen vis-à-vis the
body of knowledge that does belong to the same module as that knowledge. In any case, this is
not an entirely unreasonable conjecture. And if it is correct, it is consistent with Evans’ conten-
tion that it is of the very nature of knowledge that it not be frozen.
15. This point is clearly articulated and powerfully argued in Hacker (1999: 52-55). Hacker
plausibly attributes it to Wittgenstein.
 Conceptual Atomism and the Computational Theory of Mind

diverge from the behavior of somebody who doesn’t grasp those concepts and who is
therefore merely acting in accordance with the relevant rules.
The idea that Smith’s behavior would not thus diverge involves the idea that Smith’s
grasping the concepts twelve, addition, fifteen, and so on, doesn’t involve his grasping
any other concepts. But that idea is false. Grasping any one of those concepts involves
grasping various others. Further, the behavioral consequences of grasping one concept
are ceteris paribus different from the behavioral consequences of grasping any other
concept. Therefore, the hypothesis that Smith is really adding twelve and fifteen is rich
in predictive implications not had by the hypothesis that Smith is merely acting in ac-
cordance the rules governing the addition of those numbers. By the same token, if the
hypothesis that Smith is adding those numbers is wrong, it is extremely unlikely that
Smith’s behavior will be consistent with it for any length of time. After a certain point,
it becomes only a skeptical possibility – comparable to the possibility that there is no
external world and to the possibility that everyone besides oneself is an inanimate
zombie – that Smith wasn’t actually following arithmetical rules.
It is not hard to see the relevance of this to the debate between Chomsky and
Searle. If Smith is really following R, then he obviously grasps the concepts involved in
it. In that case, his grasping those concepts will come out in ways that are independent
of his following that particular rule. A corollary is that if Smith is not actually follow-
ing R, there will soon be a wealth of evidence showing as much.
Let us sum up. Given any rule at all, the hypothesis that someone is following that
rule is confirmationally different from the hypothesis that he is merely acting in ac-
cordance with it. A corollary is that, for any two distinct rules, the hypothesis that a
person is following the one is confirmationally different from the hypothesis that he is
following the other. Chomsky shows that there is considerable evidence for a number
of hypotheses of the form X subpersonally follows R. Searle argues that the very same
evidence confirms X is acting in accordance with R. That simply isn’t true, as we’ve seen,
and we must reject this particular argument of Searle’s.
We have seen reason to believe that Searle’s criticisms of Chomsky’s cognitive real-
ism are unfounded. But many of the views that motivate those criticisms are correct.
Given the points that we’ve developed in this section, it is possible to separate what is true
from what is false in Searle’s important but misguided “critique of cognitive reason.”16
Searle rightly holds that calculators per se are not following rules. My calculator
always displays a “13” after I punch in “6+7=ENTER.” But it doesn’t follow that my
calculator is really doing addition. All that follows is that my calculator is behaving like
something that is doing addition. Further, Searle holds (correctly, as we have seen),
that the calculator isn’t adding, at least not in virtue of its engaging in behavior of the
kind just described.
But Searle makes some false claims on the basis of these correct ones. First, he con-
tends that a thing’s acting like something that follows R isn’t non-trivial evidence that it

16. This is the title of one of the chapters of Searle (1992).

 Chapter 18.  Fodor’s third argument for conceptual atomism 

really is following R. In fact, the opposite conclusion is to be drawn. The reason we must
hold that the calculator isn’t really adding is specifically that its behavior is ultimately not
like that of something that really does add. A thing’s acting just like an actual arithmeti-
cian is surely good evidence that it is an arithmetician. (Of course, because behaviorism
is false, that evidence is defeasible.) And so far as there is evidence to suggest that calcu-
lators are not arithmeticians, that evidence lies in the fact that they do not, in the final
analysis, act very much like actual arithmeticians. To be sure, where expressions denot-
ing certain operations on small numbers are concerned, calculators and people (i.e. ac-
tual arithmeticians) do behave similarly, at least up to a point. (I say “up to a point”, be-
cause the behavior of calculators differs significantly from that of people even within
those limits.) But past that point they don’t behave similarly, as the following story may
help illustrate.
Timmy is four years old. He speaks English perfectly, and is cognitively normal for
the most part. But he is unusual in one respect. He has completely memorized his ad-
dition and multiplication tables, and he has also perfectly assimilated the algorithms
for the multiplication, division, and addition of multi-digit numbers. If asked a ques-
tion like “what is 8×9?”, he can readily produce the right answer; and if asked “what is
87×96?”, he can generate the right answer after a minute or so. But Timmy simply can-
not give the right answer to the question: “if Jones has exactly two shoes, and Brown
has exactly two shoes, then how many shoes do Smith and Brown have altogether?”
Timmy is equally incapable of answering any other comparable question.
Under these circumstances, there would be much to be said for the view that Tim-
my did not really grasp the concepts of addition, multiplication, and so on. We would
explain his ability to crank out the products and sums of multi-digit numbers by im-
puting to him a kind of meta-linguistic knowledge that simulated actual mathematical
knowledge. Timmy knows how to operate with the symbols “87×96” and “83×845.” But
he doesn’t know how to operate with the corresponding concepts.
Timmy’s behavior is very similar to that of a calculator. In fact, Timmy’s behavior
is more like that of a calculator than that of practically any other human being. At the
same time, Timmy’s behavior is not like that of someone who really grasps the con-
cepts of arithmetic. The same is therefore true of the behavior of a calculator. Searle is
entirely right to hold that calculators don’t calculate. But he is wrong to hold on this
basis that X’s acting like something that follows rule R is little or no evidence X really
is following R.
An analogy may be helpful. When we think of gravity, we tend to think of things
falling down. So when we see images of astronauts floating about in a spacecraft, we have
a tendency to think that there is no gravity in outer space. But this is obviously not the
right conclusion to draw. In certain extremely narrowly defined contexts, one of the
more obvious effects of gravity is to make things fall. But it doesn’t follow that there is no
gravity where things aren’t falling. Our tendency to transition from the astronaut is float-
ing to there is no gravity in the spaceship reflects the peculiarities of the circumstances
under which we came to acquire the concept of gravitational attraction. Given how we
 Conceptual Atomism and the Computational Theory of Mind

came to acquire that concept, we see as definitive of the presence of gravitational attrac-
tion what is in fact one of many possible expressions of its presence.
For obvious reasons, when we think of arithmetical knowledge, we tend to think
of people manipulating numerals. Our culture and educational system being what they
are, there is indeed a good correspondence between a knowledge of arithmetic and an
ability to manipulate certain expressions in certain ways. What we don’t always realize
is that somebody’s manipulating the aforementioned expressions in the aforemen-
tioned ways is evidence of arithmetical knowledge only to the extent that it correlates
with many other behaviors that don’t fit our stereotype of somebody who knows arith-
metic. If we strip away those other behaviors, leaving only the symbolic manipulations,
we are left with a faux-arithmetician, like Timmy, not an actual one. Calculators act
just like these faux-arithmeticians. So given only that calculators merely simulate ar-
ithmetical knowledge, there is no reason to deny that acting like an actual arithmeti-
cian is good (though defeasible) evidence that one really is an arithmetician.
Of course, what we just said about arithmetical rules is true of all rules. So if, as Searle
seems to grant, we act like things that follow rules involving concepts like c-command and
dominant node, then there is ipso facto good evidence that we really do follow those rules.
Before continuing, I would like to address a possible objection to the argument
just presented (it is not one that an opponent of CTM, such as Searle, would make):
You begin with some correct points, but end up with a very wrong one. Here are
the correct points. If Timmy really knows arithmetic, then he should be able to
figure out how many houses Bob has, if he knows that Bob has two more houses
than Charles and that Charles has three houses. Given only that Timmy can ma-
nipulate the right symbols in the right way, it doesn’t follow that he grasps arith-
metic; for it could be that he knows how to manipulate the symbols even though
he has no idea what they mean. In general, if a person grasps arithmetic, that
knowledge will be expressed in ways other than his or her being able to rattle off
sums and products.
So far so good. But the cracks in your argument become obvious the moment
we ask the question: Why is a person’s arithmetical knowledge expressed in these
various other abilities? Why is a person who grasps arithmetic able to make judg-
ments about the number of houses that Bob owns and about the number of shoes
that Mona bought? Because a person grasps the concepts house, shoe, buy, and so
on. But given that calculators don’t grasp these concepts, it would be absurd to
suppose that their knowledge of arithmetic would be expressed in their making
judgments involving these concepts.
Let C be some concept that Martians grasp but that human beings do not. It
would be absurd to suppose that our arithmetical knowledge would be expressed
in our making judgments involving C. Our arithmetical knowledge is expressed in
terms of concepts that are at our disposal, not in terms of those that are not. Simi-
larly, a calculator’s behavior is expressed in terms of the concepts that are at its
disposal, not in terms of concepts – e.g. boat, house, and shoe – that are not. My
 Chapter 18.  Fodor’s third argument for conceptual atomism 

calculator’s concept of addition interacts with the few other concepts at its dis-
posal at least as well your concept of addition interacts with the various concepts
at your disposal.
As you say, the behavior of a calculator is different from that of a person who
knows arithmetic. But this is precisely what we would expect, given that people
grasp various concepts that calculators do not. So you have failed to show that a
calculator’s behavior is relevantly different from that of something that really un-
derstands arithmetic. For you have failed to show that, supposing that calculators
really do know arithmetic, their behavior is anything other than what we would
expect it to be, given that they have such a limited repertoire of concepts.

This objection inadvertently exposes a reason to hold that, even after the relevant back-
ground-facts are taken into account, a calculator does not act like an actual arithmeti-
cian. Suppose that Smith knows arithmetic but that he doesn’t know any of the symbolic
conventions that (in our culture) are associated with a knowledge of arithmetic. So Smith
doesn’t know that “1” denotes 1, that “×” stands for the operation of multiplication, and
so on. For the very reasons given by the objector, it would obviously be absurd to suppose
that Smith’s knowledge of arithmetic would be expressed in his writing down “7×9=63”,
“1+12=13”, and so on. And if Smith did produce such inscriptions, then (supposing that
we didn’t revise our views as to his symbolic knowledge) we’d have to assume that they
were not expressions of his arithmetical knowledge.
It would be absurd to suppose that somebody grasped the fact that “1” denoted one,
that “×” stood for the operation of multiplication, and so on, but that he didn’t grasp the
concept of denotation. And it would be absurd to suppose that somebody grasped the
concept of denotation but that he didn’t grasp the more general concept of representa-
tion. Once it is granted that somebody knows the symbolic conventions associated with
arithmetic, it is de rigueur to grant that he also grasps various other notions – notions
that even CTM-hardliners would be unwilling to impute to calculators.
Given these points, suppose that your calculator really does grasp the concepts of
arithmetic. Unless it is granted that the calculator grasps the relevant symbolic conven-
tions, along with the concepts that would be needed to have such a grasp, it would be
absurd to suppose that the computer’s arithmetical knowledge would be expressed in its
tokening expressions such as “7×9=63” and “1+12=13.” That supposition is as absurd as
the supposition that a calculator’s arithmetical knowledge would be expressed in terms
of judgments about shoes and houses. So if any case is to be made that, after their lim-
ited conceptual backgrounds are taken into account, calculators really do act like things
that know arithmetic, it must be assumed that calculators have a rich backlog of seman-
tic knowledge. In that case, there must be some independent evidence, i.e. evidence
other than their tokening expressions of arithmetic, that they in fact grasp such con-
cepts. But there is no such evidence: there is no independent evidence that calculators
grasp any of the semantic or meta-semantic concepts that we would expect to be grasped
by a creature that knew the relevant symbolic conventions. Calculators cannot see or
hear or otherwise perceive anything that would give them any reason to believe that “1”
 Conceptual Atomism and the Computational Theory of Mind

denoted one, that “×” denoted the operation of multiplication, and so on. Apart from
the fact that they token arithmetical expressions, they do nothing that would suggest
that they have a grasp of, for example, what it is for one symbol-token to have a certain
location with respect to another symbol-token. If they have semantic knowledge, that
knowledge is completely ex nihilo, and has none of the moorings in other concepts that,
for reasons both empirical and purely logical, we would expect such knowledge to have.
In view of the relevant background facts, it is especially unlikely that calculators act like
things that actually know arithmetic. The objector’s point therefore undermines itself.
Of course, the argument just presented presupposes a non-atomistic conception of
concepts. It presupposes that, if one has a given concept (e.g. a concept of the fact that “1”
denotes the number one), one also has others (e.g. the concept of representation). And,
what is related, that argument presupposes that a concept of x cannot be realized wholly
by a causal relation to x. For, given a strictly causal conception of conception, one could
maintain that calculators grasp the relevant symbolic conventions. After all, such a con-
ception allows anything x to be grasped by anything y, no matter how conceptually im-
poverished y otherwise is, so long as x stands in the right causal relation to y; and it could
be held that the social events mediating semantic conventions stand in the right causal
relations to the activities of calculators. This is at least part of the reason that sophisti-
cated advocates of CTM, such as Fodor, accept conceptual atomism as well as a strictly
causal conception of conception. But we have seen that these doctrines are false.
There is one last point to make in connection with Searle’s view we have no good
reason to think that X really is following R, given only that X acts like something that
follows that rule. Despite everything that we just said, there is no denying that – albeit
within extremely narrow horizons – calculators do act like things that really do add,
subtract, and so on. As we’ve noted, Searle makes the correct point that calculators
don’t perform any of these operations. But having made this correct point, he over-
looks another correct point of comparable importance. Even though calculators are
not themselves performing arithmetical operations, their behavior is still appropriately
understood in terms of the performance of such operations. The calculator’s behavior
has a special connection to the concepts of arithmetic: and the connection is not mere-
ly that the calculator’s behavior is easily predicted by taking the (false) view that the
calculator is actually performing arithmetical operations. People use calculators to add
numbers. Therefore the activities of calculators are constitutive of actual cases of com-
putation, even though calculators do not themselves compute. Further, calculators are
built by people who know arithmetic and who wish to embody that knowledge in their
creation. So to the limited extent that calculators act like things that add and subtract,
the fact is to be understood in terms of the fact that something really is adding and
subtracting. Of course, that something is not the calculator itself; that something is the
human being who uses (or builds) the calculator. Nonetheless, given only that calcula-
tors don’t really compute, there is no reason to deny that a thing’s acting like something
that computes is evidence that actual computation is occurring. Contrary to what
Searle seems to hold, the fact that the calculator is acting like something that computes
 Chapter 18.  Fodor’s third argument for conceptual atomism 

is not evidentially innocuous: it does correspond to the fact that something (though
not the calculator itself) really is computing.17
Now let us consider the case of the human being whose behavior is like that of some-
thing which is following various rules concerning concepts such as initial phrase-marker
and NP-postposing. Given what we’ve said, the fact that somebody acts like something
that follows such a rule is good evidence that such rules are in fact being followed. The
rule-following to which the calculator’s behavior corresponds doesn’t have to be attrib-
uted to the calculator itself, since it can be off-loaded onto a person behind the calculator.
But nothing comparable is possible in the case of the human being who is acting like
something that follows rules involving the concept of NP-postposing, and it must there-
fore be supposed that the person himself is (sub-personally) following those rules.
Wittgenstein (1958: § 201) famously said: “no course of action could be determined
by a rule, because every course of action can be made out to accord with the rule.” Given
only that something acts as though it is following a rule, there is no reason to believe that
it really is following that rule, since there are infinitely many other rules that it could, at
least in theory, be obeying.
Wittgenstein’s point is much more innocuous than it is usually thought to be. As
we saw, the hypothesis that I am following the rule add one is confirmationally very
different from the hypothesis that I am following the rule add a million and then sub-
tract 999,999. For example, it can be established that, although I can add one, I have
extreme difficulty adding and subtracting large numbers. In that case, given the rapid-
ity with which I am generating the right numbers, it is a virtual impossibility that I am
obeying the rule add a million and then subtract 999,999.
Suppose that, on some occasion, I behave in a way that is characteristic of some-
body who is adding one. Wittgenstein is obviously right that, given only my overt be-
havior on that one occasion, it is technically possible that I am following the rule add a
million and then subtract 999,999 or that I am following no rule at all. But that possibil-
ity isn’t a significant one unless one or more radically counter-inductive assumptions
are granted. It is a counter-inductive possibility that, although I am physiologically just
like other people, my thought processes bear no resemblance to anyone else’s. It is a
counter-inductive possibility that, even though the situation just described holds, my
mental states nonetheless eventuate in behavior that isn’t appreciably different from
that of other people. It is a counter-inductive possibility that, even though the first two
counter-inductive possibilities hold, the world is otherwise sufficiently induction-
friendly for us to survive for any length of time. So to validate Wittgenstein’s view, one
has to assume the truth of a purely skeptical possibility that is embedded within an-
other purely skeptical possibility that is embedded within another skeptical possibility.
It thus doesn’t appear that Wittgenstein has done anything more than make the innocu-
ous point that nothing concerning a person’s psychological make-up is logically entailed
by (non-psychological) descriptions of his overt behavior. And, contrary to what Witt-

17. A similar argument is found in Chomsky (1998: 166-224).

 Conceptual Atomism and the Computational Theory of Mind

genstein holds, when we set aside purely skeptical possibilities, so and so’s behavior is
good evidence that he is following the rule add one, and not some other rule.
Wittgenstein is right that, given only some one act on so and so’s part, the hypoth-
esis that so and so is following some one rule (add one) is not more likely than the
hypothesis that he is following some other rule (add a million, then subtract 999,999).
Given any hypothesis that isn’t logically false (e.g. Smith’s house is a square circle), that
hypothesis is equiprobable with many hypotheses that we regard as outlandish pro-
vided that we focus on a small enough data-sample. But when we consider the data-
samples that actually motivate our judgments, it is not a significant possibility that a
bent hypothesis (e.g. Smith isn’t adding one: he is adding a million and then subtracting
999,999) is the right one.
Thus, so far as it doesn’t depend on the truth of counter-inductive, purely skeptical
possibilities, Wittgenstein’s position goes through only if it is assumed that there is an
extremely strict connection between one’s grasping a given concept and one’s acting in
a certain way. If the only evidence that I were following the rule add one, as opposed to
some bent rule (e.g. add a million and then subtract 999,999) were my acting in some
specific way on some specific occasion, then there would indeed be no reason to be-
lieve that I wasn’t following some other rule or that I wasn’t following any rule. But if I
really am following a bent rule, then there is soon going to be some independent evi-
dence of that fact.
A comparison might be helpful here. If a person is eating chicken not because he
is hungry, but because he believes that doing so will protect him from lethal galactic
rays, then there is probably going to be independent evidence of that fact. He probably
has a history of mental illness. He probably won’t obey the usual rules of decorum
while eating. He may well be wearing unusual clothes that, in his view, protect him
from the lethal rays.
Of course, we can normalize the behavioral expressions of these bent beliefs by
imputing additional bent beliefs to our subject. Maybe he believes that, unless he has
good table manners, the protective effects of chicken-consumption are neutralized.
Maybe he believes that the clothes that people ordinarily wear are the best possible
kind of protection against the lethal rays. But, if so, there will be independent evidence
for each of these bent beliefs. If that person believes that cotton fabric protects him
from lethal radiation, he will probably have other very unusual beliefs about cotton
and about radiation; and those unusual beliefs will probably leak out in some very
unusual forms of behavior. Of course, we can normalize the behavioral expressions of
these bent beliefs by epicyclically imputing yet further bent beliefs to our subject. But
barring additional epicycles, each of these imputations will disconfirmed by other be-
haviors on the part of the subject.
For these reasons, supposing that we act as if we are following rules of the kind that
Chomsky describes, the hypothesis that we really are following such rules is decidedly
more probable than the hypothesis that we are not doing so.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

Wittgenstein and the subpersonal

John McDowell (1998), Gordon Baker, and PMS Hacker (Hacker and Baker: 1980,
1984b, 1985) make it clear that, in their view, the concept of subpersonal mentation is
incoherent.18 All of these authors are self-described Wittgensteinians; and there is no
doubt that, in their antagonism towards the subpersonal, they are being entirely true
to Wittgenstein’s thought. Even though there is no passage where Wittgenstein explic-
itly discusses the concept of subpersonal mentation, he put forth two arguments that
make it clear that he would have regarded the very notion of subpersonal mentation as
incoherent. It is on the basis of those arguments, and of the general currents in Witt-
genstein’s thought of which they are distillations, that the previously mentioned au-
thors reject the concept of subpersonal thought. These arguments are now known as
the “private language” and “rule-following” arguments. Let us now consider these ar-
guments, starting with the former.19

18. Baker and Hacker usually work together. Their books are usually co-authored.
19. My reconstructions of Wittgenstein’s arguments owe much to Kripke’s (1982) reconstruc-
tions of them. The accuracy of Kripke’s reconstructions of both arguments, especially the rule-
following argument, was strongly contested by Hacker and Baker (1984). According to the latter,
Wittgenstein is saying the exact opposite of what Kripke imputes to him. According to Kripke’s
account, Wittgenstein is advocating a kind of nihilist-behavioristic account of meaning-gras-
ping. According to Baker and Hacker, Wittgenstein takes it for granted that such an account is
false, and his analysis is meant to be a reductio ad absurdum, i.e. Wittgenstein is trying to show
that certain views must be rejected because they lead to skepticism, behaviorism and other forms
of reductionism with regard to the grasping of rules and, more generally, of meanings. Many
philosophers (e.g. McGinn 2002: 146-148) take the Baker-Hacker critique of Kripke’s recons-
truction of Wittgenstein to be correct.
In light of this, what right do I have to accept a Kripkean understanding of Wittgenstein’s
arguments? First of all, Kripke’s arguments correspond very well to what people have taken
Wittgenstein to mean. It may not be clear what Wittgenstein intended to say. But Kripke has
succeeded in articulating the doctrines that (whatever his intentions were) Wittgenstein be-
queathed to philosophy.
I am not a Wittgenstein scholar. But, so far as my untutored intuitions are able to judge, the
Kripkean account certainly seems to fit Wittgenstein’s prose. Baker and Hacker may be right to
say that, so far as his statements correspond to Kripke’s reconstruction, Wittgenstein was being
ironic or otherwise indirect. But there is nothing in Wittgenstein’s prose that makes it obvious
(to neophytes like myself) that Wittgenstein is saying anything other than what he means. Also,
in other writings of Wittgenstein (e.g. Wittgenstein 1973, 1983), one finds what seem to be non-
ironic affirmations of the doctrines that Kripke imputes to him. Most importantly, it is not ob-
vious (to me) that Kripke’s account is unfaithful to what philosophy has taken from Wittgens-
tein’s writings.
There is another relevant fact. The doctrines imputed to Wittgenstein by Kripke are extremely
clear (though I will argue that they are false), and thus form a good platform for philosophical
discussion. But the doctrines ascribed to Wittgenstein by Hacker and Baker are extremely obscure.
 Conceptual Atomism and the Computational Theory of Mind

The Private-language Argument20

Wittgenstein’s so-called “private language” argument is meant to show that there can-
not be a private language that is not a mere translation of a public language.
Let me say what is meant by the italicized qualification. Obviously a person could
create a private code for himself, as the following bit of fiction shows. I want to keep a
diary. I know that my housemates will read it. So I invent a private code that they won’t
understand. I replace each letter with the cube of the number giving the place of that
letter in the alphabet. So instead of “b”, I write “8”; instead of writing “e” I write “125”;
and so on. This private code is obviously parasitic on the existence of a public language.
Wittgenstein doesn’t deny that there can be private languages in this sense. Wittgen-
stein denies that there can be private languages that are not thus parasitic.
Before we give Wittgenstein’s argument, let us say a word on why it is thought to
be important. It is widely thought that if there is subpersonal mentation, it is mediated
by some kind of language. This view is held both by advocates and opponents of cogni-
tive science. If this is right, then there can be no subpersonal mentation, if Wittgen-
stein’s argument is cogent (Fodor 1975, Dennett 1978: 91–95, Pylyshin 1984, Fodor
and Pylyshin 1988). Advocates of cognitive science say that there is subpersonal men-
tation and thus some kind of private language, and they say that the Private Language
Argument must therefore contain some kind of error (Fodor 1975).21 Opponents of
cognitive science say that the Private Language is cogent and that, consequently, there
is no private language and thus no subpersonal mentation. But we will see that, if it
exists, subpersonal thought need not, and indeed cannot (at least not generally), be
mediated by a language or even by anything at all language-like. So even if Wittgen-
stein’s argument is cogent, that fact has no bearing on the viability of cognitive science.
But it is obviously worth our while to examine Wittgenstein’s argument, and that is
what we will now do.
Here is my reconstruction of the Private Language Argument:
Smith doesn’t speak a language. But he decides that he wishes to record his
thoughts, and thus decides to construct a language for himself. He assigns mean-
ings to a number of expressions. In particular, he stipulates that:

Before they could be assessed for truth or falsehood, those doctrines would have to be precisified.
This would be a time-consuming task, and its results would inevitably be disputed.
For these reasons, I am going to take it for granted that Kripke’s (1982) reconstruction is at
least approximately correct.
20. The Private Language argument is given in Wittgenstein (1958: §§ 234-334).
21. At the same time, those cognitive scientists almost never say what, exactly, the error in
question is supposed to be. Other philosophers, e.g. A.J. Ayer (1968), have identified at least
some of the errors in that argument. But those same philosophers are typically unreceptive to
the idea that there is subpersonal, or even personal but subconscious, ideation.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

(*) “glump berka der”

is to mean:

(**) Today it snowed.

Smith does all of this on a Monday. That same day, before he forgets any of the
semantic rules that he has created, Smith writes (*) in his diary. (His intention, of
course, is to write down a sentence that means: today it snowed.) On Wednesday,
Smith looks at what he wrote on Monday. In the intervening two days, he has
forgotten what (*) means. But he thinks that he remembers what (*) means; he is
convinced that (*) means:

(***) yesterday it was sunny.

Here we can take one of two positions. On the one hand, we can say that Smith
misunderstood the inscription of (*) that he was looking at: (So (*) really means:
today it snowed. Since Smith thinks otherwise, he is wrong.) On the other hand,
we can say that (*) now really does mean (***). But it is not hard to show that the
first option is untenable and that we must therefore accept the second. Suppose
that, one day, every English-speaker in existence suddenly came to believe that the
sound “blue” meant red and that the sound “red” meant blue. (To simplify discus-
sion, let us suppose that any books or records containing any indication of the
previous usages of those terms were suddenly destroyed.) In that case, “blue”
would mean red and “red” would mean blue. Of course, if I alone believe that
“blue” means red and “red” means blue, I will be wrong. But I would be wrong only
because there would be millions of other people whose usage of those terms didn’t
correspond to my belief.
Given that Smith is the only speaker of his language, his deciding that (*) actu-
ally means (***) is comparable to the case where every English speaker suddenly
takes “red” to mean blue; it is not comparable to the case where I alone (mistakenly)
reverse the meanings of “red” and “blue.” Smith’s usage of a given expression (of his
language) sets the standard. It would therefore be absurd to say that Smith has mis-
understood an expression of that language. That would be like saying that the word
“probably” really means demonstrably (i.e. capable of being shown to be true on strict-
ly deductive grounds), since that is what “probably” meant five hundred years ago,
and that every English speaker alive today is therefore wrong to take “probably” to
mean having more than a 50% chance of being true.
If this is right, then Smith cannot possibly be wrong as to what an expression of
his language means. Any expression of his language means what he thinks it means.
In that case, there is no wrong way to use any expression of that language. (Of course,
matters would be different if other people started using those expressions. But, by
hypothesis, the language in question is spoken only by one person.) If there is no
wrong way to use any expression of a language, then none of those expressions has
any meaning. “Snow is green” is meaningful because there are right and wrong ways
to use it. If somebody learning English as a second language says “snow is green”,
 Conceptual Atomism and the Computational Theory of Mind

obviously intending to say snow is not green, we can correct him: we can say that he
used that sentence wrongly. But this is not an option in Smith’s case.
A private language is one whose expressions cannot possibly be used wrongly,
and it is therefore one whose expressions don’t have any meaning at all. Of course,
it is absurd to suppose that there could exist a language whose expressions were
categorically meaningless, since a language just is an assignment of meanings to
expressions. Therefore, it is absurd to suppose that there could be a private lan-
guage. Thus, cognitive science is incoherent in so far as it demands that there be
such a language.

This argument involves a failure to make three distinctions: first, the distinction between
non-identical, but historically connected, languages; second, the distinction between the
consequences of breaking a semantic rule and the social consequences of those conse-
quences; and, third, the distinction between a condition’s being constitutive of a language
and a condition’s being causally necessary for the continued existence of a language.
The language that we speak is historically connected with the language that Chau-
cer spoke, and they are both referred to as “English.” But the set of semantic rules
constitutive of the one language obviously doesn’t coincide with the corresponding set
for the other language. Strictly speaking, we are dealing with two different languages,
albeit ones that are similar and also genetically related.22
So supposing that (*) has changed its meaning, it follows the language Smith was
using on Monday is distinct from the language. Let LM be Smith’s Monday-language,
and let LW be Smith’s Wednesday-language. It may be true that (*) means whatever
Smith thinks it means, given that he is the only person who gives meaning to that ex-
pression. But it doesn’t follow that the view he has on Wednesday as to what (*) meant
on Monday is necessarily correct. So it doesn’t follow that Smith is always right in his
views as to what LM’s expressions mean. What follows is that, on Wednesday, Smith is
right to as to what is meant by the LW-homonyms of LM’s expressions. It may be true
that, for any time t, Smith is necessarily right as to what is meant by the expressions of
Lt. But it doesn’t follow that, for any later time t*, Smith is right at t* as to what was
meant by the expressions composing Lt. All that follows from Wittgenstein’s argument
is that Smith is right about the meanings of the expressions composing Lt*, where Lt* is
a language that sounds like Lt but may be characterized by different rules and may thus
not be identical with Lt.
Wittgenstein equivocates on the term “language.” Sometimes he uses it to refer to
(synchronic) sets of semantic rules, and sometimes he uses it to refer to (diachronic) se-
ries of such sets. Wittgenstein’s argument implodes when this equivocation is exposed.

22. Of course, one can use the term “language” to refer, not to a set of semantic rules, but to a
certain kind of sequence of such sets. By that definition, our English is identical with Chaucer’s
English. But in this context it is obviously preferable not to take the word “language” in this way.
Also, if that word is defined in the way just proposed, then what are obviously different langua-
ges, e.g. Latin and Spanish, end up being falsely identified with each other.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

In Wittgenstein’s defense, it might be said that our argument has merely evaded the
problem, not solved it. For it may be said that LM is so useless, so incapable of transmitting
information from Monday-Smith to Wednesday-Smith, that it doesn’t deserve to be de-
scribed as a “language.”
But whether LM is thus useless is entirely an empirical question, and it is one whose
resolution doesn’t favor Wittgenstein’s argument. Clearly Smith could use LM to relay
messages to himself. Suppose that, on Friday, Smith remembers the semantic rules that
he laid down on Monday and that, on that basis, he decodes his diary-entry from that
day. In that case, LM did do its job – it did relay information from Monday-Smith to
Friday-Smith.
Of course, LM’s being able to function in that way presupposes that Smith remem-
bers the relevant rules. But, as Ayer (1968) points out, that is true of any language at all.
If I forget the semantic rules of English, then English-utterances won’t mean anything
to me. In general, such utterances are meaningful only in so far as people remember
the relevant semantic rules. Of course, if I forget what some word means, other people
are available to remind me of its meaning. (That is why “furtive” is meaningful, even
though many don’t know what it means.) But that doesn’t point to some ontological
difference between English and LM. Where both languages are concerned, semantic
rules are sustained by human memory and nothing besides. If every English speaker
suddenly forgot how to speak English, then that language would cease to exist (except
in the irrelevant sense in which Latin and Sanskrit still exist).
Of course, because English is spoken by millions of people, whereas LM is only used
by one, the former can survive the faulty memory of any one of its users, whereas the
latter cannot. But, to echo what we said a moment ago, this doesn’t indicate some onto-
logical difference English and LM. In each case, it is human memory, and human mem-
ory alone, that endows expressions with content. In the one case, those memories are
duplicated many times over and there are therefore more back-up mechanisms waiting
to step into the breach left by any one person’s faulty memory. But that shows only that
the continued existence of English is causally better guaranteed than that of LM.
During the time that it stands, a house that stands for a week is no less a house
than one that stands for a year. The difference between English and LM is that, for what
are basically sociological reasons, the former is more enduring than the latter. It isn’t
that the preconditions for meaning have been met in the one case but not the other.
This is clearly indicated by the fact, previously noted, that there is nothing absurd in
supposing that Smith could use LM to communicate with himself.
Wittgenstein’s argument conflates the concept of a condition’s being causally neces-
sary for the existence of a language and a condition’s being necessary for there being a
language (i.e. a set of meaningful expressions) to begin with. A related point – I will leave
it open whether it is equivalent with, or merely similar to, the one just made – Wittgen-
 Conceptual Atomism and the Computational Theory of Mind

stein’s argument seems to involve a conflation of the concept of a language’s continuing to

exist, on the one hand, and its coming into existence, on the other. 23

Consequences of the misuse of expressions versus consequences

of such consequences

Wittgenstein points out that if somebody misuses an expression belonging to a public

language, there are consequences. If I tell my doctor “I feel enervated”, when what I
mean is I feel overly energized, he will give me the wrong medication (assuming that he
knows what “enervated” means and that he believes that I know what it means). Witt-
genstein holds that nothing comparable happens in Smith’s case. Wittgenstein con-
cludes that the concept of a rule doesn’t apply to Smith’s situation. A rule that can be
broken with impunity is no rule at all. Since Smith has carte blanche to break the so-
called rules of LM, it follows that there are no such rules.
First of all, it is straightforwardly false to say Smith won’t be penalized for misunder-
standing the expressions of his language. Because of such a misunderstanding, he may end
up eating the poisonous berries, instead of the good ones, since he then thinks that “blub-
bo” means good berry instead of poisonous berry. But even if we set this point aside, Witt-
genstein’s argument confuses the consequences of breaking a semantic rule with the conse-
quences of those consequences. Suppose that Jones promises to give me a million dollars if
I manage to solve a difficult math problem. The solution to the problem is 712. I believe
that the solution is 418. But I don’t speak English well, and I say “the solution is 712”, think-
ing that I have said the solution is 418. Jones gives me a million dollars.
Here I broke the rules of English-semantics. (I misused an expression.) Because I
broke the rules of English semantics, Jones took me to mean the solution is 712 when
what I meant was: the solution is 418. So by itself my breaking the semantic rules of
English had a negative consequence: I was misunderstood. In its turn, that conse-
quence, negative though it was, had another consequence; and the goodness of the
second consequence obviously outweighed the badness of the first. But my being mis-
understood was, by itself, a negative consequence of my linguistic incompetence.
A brief digression through meta-ethics may clarify the principle at work here. Bad
consequences that are outweighed by good ones do not, in virtue of that fact, cease to
be bad. Suppose that I catch a cold and thus cannot make my flight. The plane that I
would have flown on crashes. By itself, my catching a cold is a bad thing: surely physical
illness is a bad thing. That bad thing had a consequence which, for its part, was very
positive. But it would be absurd to say that, because it had a good consequence, my
being physically ill was good in and of itself. Similarly, it would be absurd to say that in
and of itself my being misunderstood by Jones was a good thing. Being misunderstood

23. Ayer (1968) provides a similar criticism of the Private Language argument.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

is a bad thing, even though, like practically any bad thing, it may happen to have a
good consequence.
If I misunderstand a sentence-token of my private language, a bad thing has hap-
pened: one person failed to make himself understood to another – Monday-JM failed
to make himself understood to Wednesday-JM. For many reasons, Monday-JM may
not be given any penalties over and above being misunderstood for his failing to make
himself understood to Wednesday-JM. (I say “may not” because, as previously dis-
cussed, JM’s misunderstanding may lead him to eat the wrong berries.) There will be
no court-hearings, no community-service, and no expulsion from the Rotary club. But
there will be a penalty: in being misunderstood, Monday-JM was ipso facto penalized.
So Wittgenstein is quite wrong to say that the rules of a private-language can be broken
with impunity.
If Smith misinterprets an expression of LM, there is a negative consequence: Smith
has not made himself understood to himself. (Monday-Smith failed to relay a message
to Wednesday-Smith.) Of course, that consequence will not, in its turn, have any social
consequence. (though it may have health-consequences, as we discussed). But Smith
did incur a significant penalty for misinterpreting an utterance of (*): there didn’t have
to be any second (or third…) party for this to happen. Wittgenstein may (or may not)
be right to say that there are no rules where everything is permitted. But, as we just saw,
not everything is permitted where LM is concerned.
Also, it doesn’t seem true that there are rules only where there is some kind of en-
forcement. There aren’t necessarily any penalties for mis-adding numbers. (As we just
saw, mathematical incompetence may be rewarded.) Arguably, certain kinds of social
rules cannot exist where there is no possibility of punishment. (If there is no penalty
for dumping one’s garbage in a certain lake, then – so one might argue – there is no
rule prohibiting it. But even this is false. Breaking rules is what causes people to be
punished. My driving too fast is what causes me to be given a ticket. If we say that,
where there is no punishment, there are ipso facto no rules, then we are saying that the
occurrence of punishments is constitutive of the existence of rules. Constitutive rela-
tions are not causal relations, since the latter, but not the former, hold between distinct
states of affairs. So if we say that there are no rules where there is no punishment, we
are denying the obvious truth that the breaking of rules can bring about punishment.
Thus, there can be unenforced rules. This by itself undermines Wittgenstein’s argu-
ment. But, as we will now see, Wittgenstein’s argument fails even if we grant his as-
sumption that there must be punishment for there to be rules.) But Wittgenstein can-
not use that fact to show that the existence of linguistic rules presupposes a plurality of
people. After all, what he is trying to show is precisely that linguistic rules fall into the
category of social rules (or, at any rate, of those social rules whose existence presup-
poses the possibility of punishment).24

24. See my article “Can one grasp propositions without knowing a language?” (Kuczynski
2005b) for an argument similar to that just given.
 Conceptual Atomism and the Computational Theory of Mind

Why it doesn’t matter whether the Private Language argument is cogent

We have seen that the Private Language is fallacious and therefore proves nothing. But
why is it thought that, if it were cogent, it would warrant the rejection of the idea that
there is subpersonal mentation? Why would subpersonal mentation necessarily in-
volve a language?
As we’ve seen, Fodor thinks that all thinking, and therefore all subpersonal think-
ing, involves a language. His view is that, if subpersonal mentation were to exist, it
would necessarily involve some kind of internal code, by means of which different
cognitive modules would communicate with one another. Many cognitive scientists
agree with Fodor about this.25
But contrary to what both some cognitive scientists and Wittgensteinians think,
the non-existence of a private language doesn’t have any bearing on the question
whether there is subpersonal thought or, therefore, on the question whether the theo-
ries posited by Chomsky, Marr, and other cognitive scientists are correct. Those theo-
ries demand that there be subpersonal exchanges of information among different cog-
nitive subsystems. But it doesn’t follow that those exchanges involve a language. Given
only that information-exchanges between people typically involve a language, it doesn’t
follow that the same is true of information-exchanges within a single person.
Computationalists often seem to say that any representational system is ipso facto
a language.26 But if that is correct, then the thesis that we think in language is reduced
to the triviality that we have thoughts. So to the extent that it is not trivial to suppose
that there is a language of thought, there must be non-linguistic modes of exchanging
and encoding information. So cognitive science must hold that not all mental opera-
tions are linguistic. A consequence is that, even if it were cogent, the Private Language
Argument would not necessarily have any bearing on the viability of that discipline.
Cognitive scientists are actually conceding too much to Wittgenstein. Wittgen-
stein held that language is inherently social, and he held that thought is inherently
linguistic. For this reason, he held that thought is inherently social. Cognitive scientists
hold that thought is not inherently social, but they often grant Wittgenstein’s assump-
tion that thought is inherently linguistic. They thus choose to reject Wittgenstein’s the-
sis that language is inherently social. (See Fodor 1975, especially the chapter titled
“Why there must be a private language.”) But Wittgenstein is wrong to hold thought is
linguistic; and in accepting this false view of Wittgenstein’s, cognitive scientists are
playing into his hands more than they have to.
We may conclude that Wittgenstein’s Private-language Argument is fallacious and
also that, even if it were cogent, that fact would have no bearing on whether theories
such as Chomsky’s are correct.

25. For example, see Dennett (1978: 39-52, 90-108).

26. See Fodor (1975).
 Chapter 18.  Fodor’s third argument for conceptual atomism 

The rule-following argument27

Consider the rule:

(S) You must stop at stop-signs.
Smith is driving along. He sees a stop-sign. He knows of S and therefore stops his car.
One analysis of this situation is as follows. S is some kind of abstract or hyper-real
object, and Smith’s knowing about S consists in his beholding this object. He applies
the information given to him in this platonic insight to his current situation. Because
he applies this information correctly, he decides to stop his car.
Here is what Wittgenstein says. For the sake of argument, suppose that the analysis
just given is correct. If Smith applies his knowledge of the aforementioned hyper-real
object in a haphazard or arbitrary manner, he will make the wrong decision: he will
decide not to stop. Smith must therefore apply his knowledge of this object in a princi-
pled manner. This means that he must follow some principle or rule S1 that says how
to apply S. But by the same logic, he must then follow some rule S2 that says how to
follow S1. We have generated a vicious regress.
Consider the function F(x)=x2. You haven’t considered each of the ordered pairs
constitutive of this function. For example, you have not (let us suppose) considered
F(937). How would you find out the value of F(937)? One is tempted to answer this by
saying: “To grasp F(x)=x2 is to be aware of some platonic entity, and one figures out the
value of F(937) through a principled application of the information embodied in this
awareness.” In other words, one applies the first principle by knowing some second
principle. But then the question that we asked in connection with the first principle
arises in connection with the second, and we are no closer to explaining how you are
able to produce the right answer.

27. The Rule-following argument is given in Wittgenstein (1958: §§ 138-238, 1983: §§ 5-111).
Before we present Wittgenstein’s argument (or, in any case, my reconstruction of it), a preliminary
point is in order. There are different kinds of rules. The function F(x)=x2 can be described as a
“rule”, and so can the law requiring one to stop one’s vehicle at certain kinds of signs (red, octagonal
signs on street-corners, with the word STOP on them). The second describes an obligation, and
thus has a normative component. The first has no such component and merely describes an abs-
tract pairing of numbers. These two “rules” are thus in very different categories. In fact, given that
it applies to such different things, the word “rule” is arguably ambiguous.
In his rule-following argument, Wittgenstein appears not to clearly distinguish rules like F(x)=x2
from rules like you must stop at stop-signs. In fact, it seems as though part of what he is arguing is that
there is no distinction and that the first class of rules is a sub-set of the second. In other words, Wit-
tgenstein sometimes seems to be saying that rules of the first kind are ultimately social rules, and are
thus in the same category as rules of the second kind. In any case, in keeping with Wittgenstein’s own
usage, we will use the word “rule” indiscriminately in our reconstruction of Wittgenstein’s argument,
and will not worry about any pluralities of meaning that might characterize it.
 Conceptual Atomism and the Computational Theory of Mind

Wittgenstein then seems to suggest that principled behavior isn’t to be explained

in terms of grasping anything at all. One sees; one reacts. There is no intervening grasp
of platonic entities. One just does it.
Wittgenstein makes it clear that, in his view, there is a distinction between principled
and unprincipled behavior. But he seems to think that the difference lies not in anything
that is internal to the subject’s psyche. As Wittgenstein puts it, if God were allowed only
to see inside people’s minds, He wouldn’t always be able to distinguish between some-
body who happened to make a correct guess as to the value of F(937) and somebody who
genuinely figured that value out. Wittgenstein insists that the difference lies not in the
individuals themselves, but in the way in which they are embedded in their respective
social environments. Wittgenstein is thus a content-externalist.
One conclusion that is drawn from the rule-following argument is that the ability to
follow rules – to apply functions, execute algorithms, and generally think and act in a prin-
cipled manner – presupposes that one is embedded in a certain way in certain social prac-
tices – in particular, those involving the use of symbols. Of course, a pre-condition for one’s
being thus embedded is that one have sense-perceptions, be able to interact with others,
and so forth. So if the use of symbols is a prerequisite for the following of rules, then it
makes no sense to say that cognitive achievements like language-learning and perception-
generation presuppose that one already has the ability to follow rules. So if the rule-follow-
ing argument is cogent, then the very idea of a non-social or pre-social rule-follower is
incoherent. And in that case, then the hypotheses put forth by Chomsky and Marr are in-
coherent, since the instances of rule-following posited by those theories are pre-social.

Evaluating the rule-following argument

Wittgenstein’s argument involves a failure to register distinctions of scope. There isn’t

a distinction between grasping (S) and knowing how to apply it; and knowing how to
apply (S) doesn’t involve grasping some second principle.
A correct analysis of the matter lies in the following principle:
(S*) To grasp S just is to know of anything x, such that one believes x to have certain
properties (red, octagonal…), that one is ipso facto supposed to stop at x.
Before we continue, we must distinguish (S*) from two other statements:
(S**) To grasp S just is to know that, given anything x having certain properties
(red, octagonal…), one is ipso facto supposed to stop at x.
and
(S***) Given anything x having certain properties (red, octagonal…), to grasp S just
is to know that one is ipso facto supposed to stop at x.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

If (S***) is right, then given every stop-sign x in the world, anyone who grasps (S) is
aware of x individually, i.e. any such person knows of each stop-sign x in existence that
one is supposed to stop at x specifically. Since that is false, so is (S***).
(S**) is false for the reason that Wittgenstein himself gives: (S**) explains the
grasping of principle (S) in terms of the grasping of some second principle (given any-
thing x that is red, octagonal, and so on, one is supposed to stop at x). So, just as Wittgen-
stein argues, (S**) merely passes the buck, leading to a vicious regress.
But Wittgenstein’s argument doesn’t apply to (S*); and (S*) doesn’t falsely presup-
pose that each person who knows (S) is acquainted with every individual stop-sign.
(S*) is to the effect that a person’s knowing (S) consists in his knowing of anything x
that he encounters (or otherwise considers) and that he believes to be red, octagonal,
and so on, that x is, in virtue of having those very properties, such that one is to stop at
it. Given that (S*) is true, a knowledge of how to apply S is built into that same knowl-
edge. There is no need to invoke knowledge of a second principle.
A short story will help clarify one possibly obscure feature of this analysis. Smith is
driving along and he sees what is in fact a stop-sign. But he doesn’t realize that he is see-
ing a stop-sign, the reason being that, because of unusual lighting conditions, his visual
perception leads him to believe that what he is seeing is round and yellow, as opposed to
red and octagonal. So Smith doesn’t stop. Does this mean that Smith doesn’t grasp S?
Of course not. To grasp S is for it to be the case that, for any object x that one con-
siders, and that one believes to have the relevant properties (red, octagonal…), one
knows that, in virtue of (supposedly) having those properties, x is such that one must
stop at it. Grasping S obviously doesn’t involve being infallible, or even reliable, at dis-
tinguishing things that are stop-signs from things that are not. Grasping S consists in
its being the case that when one believes (whether truly or falsely) of an object that it
has certain properties, one believes that one is ipso facto to stop at it.

An alternative formulation of our analysis

Here is a different way of articulating our thesis. This re-articulation is not needless
repetition: for it will help us map what we have said about relatively simple rules, like
(S), onto more complex ones, like F(x)=x2.
Grasping S consists in its being the case that, if one believes x to have certain prop-
erties (red, octagonal…), and if one believes y to be a case of stopping at x (at a certain
time after beholding x), then one believes that, in virtue of those facts, the pair <x,y> is
to be assigned to some class C; and grasping S consists in its also being the case that, if
one believes x to have certain properties (red, octagonal…), and if one believes y to be
a case of not stopping at x, then one believes that, in virtue of those facts, the pair <x,y>
is to be assigned to the complement of C. So grasping S consists in one’s believing that,
for some class C, <x,y> falls into C if y is a case of stopping at x, where x is an object that
is red, octagonal, and so on, and that otherwise <x,y> falls into the complement of C.
 Conceptual Atomism and the Computational Theory of Mind

Thus, one’s grasping S consists in two conditions being met:

(*) One believes, of any two cases that one considers where (so one thinks) some-
body stops his vehicle at a certain kind of object, that those two cases are ipso
facto to be co-classified (i.e. put into some one class).
(**) One believes, of any two cases that one considers where (so one thinks) some-
body has failed to stop his vehicle at a certain kind of object, that those two
instances are ipso facto to be denied membership in the class just mentioned
(i.e. that they are both to be put into the complement of that class).
Notice that each of these conditions can be met, where a given person is concerned,
even if that person is incompetent in the way of distinguishing things that are stop-
signs from things that are not. But notice also that, if a person’s grasping S consists in
those two conditions being satisfied in that person’s case, then there is an internal rela-
tionship between, on the one hand, grasping S and, on the other hand, knowing that
one is to stop one’s car at the red, octagonal object in front of one. There is no need to
apply one’s grasp of S. There is no need to invoke some second rule to see how (if at all)
the principle bears on a particular situation.

Extending our analysis to more difficult cases

These points map onto the mathematical case. (In this context, the term “square” will
be used in its arithmetical sense – as in “16 squares 4”, not “that table is square.”) A
preliminary point will help make it clear why this case parallels the one just discussed.
First of all, suppose that Smith grasps the concept of a square. Further, Smith believes
(correctly) that 4 squares 2 and that 16 squares 4. Smith also believes (correctly) that 5
does not square 3 and that 18 does not square 7. In virtue of having these beliefs, Smith
believes that the pairs <4,2> and <16,4> fall into some class C such that the pairs <5,3>
and <18,7> fall outside of C. Grasping the concept of a square undeniably involves be-
lieving that, if two ordered pairs both have certain properties, they are ipso facto to be
put into some one class, and also that if two such pairs lack those properties, they are
both ipso facto to be denied membership in that same class.
Given this, I propose that one’s grasping the concept of a square consists in two
conditions being met:
(*) One believes, of any two ordered pairs that one considers to have certain
properties, that those two pairs are ipso facto to be put into some one class.
(**) One believes, of any two ordered pairs that one considers to lack those same
properties, that those two pairs are ipso facto to be denied membership in that
same class.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

Supposing that this analysis is right, one’s grasping the concept of a square doesn’t in-
volve one’s knowing exactly which ordered pairs have the relevant properties (just as
grasping S doesn’t involve knowing which objects are stop-signs or, therefore, which
cases of car-stoppings are cases of somebody’s stopping at a stop-sign). Further, sup-
posing that our analysis is right, there is no need to apply one’s grasp of the concept of
a square. There is no need to invoke some second rule – or therefore a third or a
fourth….— to see how that principle bears on a particular ordered pair. If one believes
that a given pair has the relevant properties, then one ipso facto believes it to fall into a
certain privileged class.
There is an obvious objection to this:
For a pair <x,y> to have the relevant properties just is for y to square x. So in order
for somebody to believe that a pair has the relevant properties, he must already
grasp the concept of a square.

To see why this is not the case, we need only remember what we said earlier about the
non-conceptual basis of conceptual knowledge. You hear a song on the radio. You know
that it is an instance of rock ‘n’ roll – not jazz, not classical, not hip-hop. You classify that
song as rock ‘n’ roll on the basis of its properties – its harmonic and melodic modalities,
its rhythms, and so on. Each of these properties can be characterized independently of
the concept of rock ‘n’ roll, and you can grasp any one of these properties without hav-
ing that concept. If Bach were to come back from the dead one day and hear that song,
he would know that it had those melodic, harmonic, and rhythmic properties. But he
would not, at least not immediately, have the concept of rock ‘n’ roll.
If a piece falls under the concept rock ‘n’ roll, that is in virtue of its having proper-
ties that are ultimately not to be understood in terms of its falling under that, or any
other, concept. A related point is that conceptual knowledge supervenes on non-con-
ceptual knowledge. Your knowing that a certain piece falls under the concept rock ‘n’
roll is ultimately to be understood in terms of your grasping information that is not to
be understood in terms of that piece’s falling under that, or any other, concept.
In general, conceptual facts supervene on non-conceptual facts, and conceptual
knowledge supervenes on non-conceptual knowledge. Conceptual truths must be seen as
articulations of non-conceptual truths, and conceptual knowledge is an articulation of
non-conceptual knowledge. If one grasps a concept – be it the concept of rock ‘n’ roll or
of exponentiation – one grasps it in terms of content that is ultimately non-conceptual.
The operative word here is “ultimately.” It is not hard to see the non-conceptual
basis for a truth of the form that song is an instance of rock ‘n’ roll. One hears the sounds
– the melodies and harmonies, and the timbres of the instruments that generate them
– and one superimposes a judgment (that is rock ‘n’ roll) on that mass of raw, non-con-
ceptual information. But some judgments are often at many removes from their non-
conceptual bases. The concept of exponentiation is grasped in terms of that of multi-
plication. The concept of multiplication is grasped in terms of that of addition. The
concept of addition is grasped in terms of the successor-relation. From a mathematical
 Conceptual Atomism and the Computational Theory of Mind

viewpoint, the successor relation seems to be an ultimate and indefinable concept.

(Set-theorists would disagree. But they take other concepts as ultimate and indefina-
ble, e.g. the concept of set-membership. Supposing that set-theorists are right, every-
thing we are saying about the successor-relation is true mutatis mutandis of that other
concept, whatever it may be.) But even though the successor-relation is a primitive
concept, it isn’t a primitive piece of information. Non-primitive concepts are conceptu-
alizations of other conceptualizations. Primitive concepts are conceptualizations of
non-conceptual information. But primitive concepts are not primitive information,
and cognitive manipulations of primitive concepts are manipulations of non-primi-
tive, but also non-conceptual, units of information. Of course, these points need to be
both clarified and substantiated. (Some of the needed clarification and substantiation
will be given in Chapters 21–23.) But if they are even approximately correct, they break
the vicious circle described by the objector.

Summary of our critique of the Rule-following Argument

Applying a concept doesn’t involve knowing of each individual object, or even of each
individual object that one considers, whether it falls under that concept or not. (Witt-
genstein is obviously right about that.) But applying a concept C doesn’t involve grasp-
ing some second concept C* such that C* is a concept of how to apply C. To grasp C is
to believe, of any two objects x and y that one considers, that x and y are to be put into
some one class if they share certain properties, and are to be denied membership in
that same class if they lack those same properties. Knowledge of a concept is thus
knowledge of a conditional proposition having the form:
(*) If x and y share property P, then (but not otherwise) they are to be co-classified.
For it to be unclear “how to apply” a concept C is for it to be unclear whether the object
(or situation) in question satisfies the antecedent of (*) (or, more precisely, of the spe-
cialization of (*) that gives the content of one’s concept of C). There is never a gap be-
tween grasping a concept and knowing how to apply it. But there is sometimes a gap
between grasping a concept and knowing whether to apply it. Given a grasp of a con-
cept C, one ipso facto knows of any object x satisfying the relevant boundary condi-
tions whether C(x) is true or not. But one often doesn’t know whether those boundary-
conditions obtain.
Since there is no gap between grasping a concept and knowing how to apply it,
Wittgenstein’s regress is blocked. Of course, there is a gap between grasping a concept
and knowing whether some object meets the relevant boundary conditions. You can
grasp the rule stop your car at signs having certain properties (red, octagonal…) without
seeing that the sign in front of you has those properties. (Maybe you aren’t wearing
your glasses; maybe you are color-blind.) But that doesn’t mean that you have to figure
out how to apply your knowledge of that law to know what you should do if it should
 Chapter 18.  Fodor’s third argument for conceptual atomism 

turn out that there is a sign with those properties in front of you. Knowledge of what
to do, in the event that there is such a sign, is built into your grasp of that concept. It is
obviously irrelevant that you do not know, for each existing object (or even for every
object, whether actual or merely imaginary, that you consider), whether it is red, oc-
tagonal, and so on. For exactly similar reasons, you can grasp the rule F(x)=x2 without
knowing of each pair of numbers, or even of each pair of numbers that you consider,
whether one of them squares the other. But, contrary to what Wittgenstein appears to
argue, this doesn’t mean that some kind of vicious regressiveness is embedded in the
concept of a concept. It corresponds only to the innocuous fact that it isn’t always obvi-
ous whether a given object satisfies the antecedent of a conditional proposition.

The personal unconscious

The concept of subpersonal mentation has proven to be a fruitful one, and we have
seen why the philosophical arguments advanced against it are less than probative. This
might be a good place to discuss the philosophical arguments that have been directed
against the concept of unconscious mentation generally. The latter arguments are, I
believe, more easily disposed of than the former.
Here is the most significant and influential argument against the notion that there
can exist unconscious mentation28:
It would make no sense to say that desks were not physical objects. A desk is a
paradigm case of such an object. Leaving aside cases of insincerity and linguistic
incompetence, anyone who said “desks are not physical objects” would be propos-
ing a redefinition of the term “physical object.”
Somebody who says that there are unconscious mental states is producing a
comparable redefinition of the term “mental state.”29 Conscious states are our
paradigms of mentality. To think of something as being a mental entity is to think
of it as being relevantly like those entities that form our paradigms of mentality
and, therefore, as being relevantly like our conscious experiences. Thus, to think
of something as being mental is to think of it as being identical with, or relevantly

28. This argument is found in Searle (1992: Chapter 7). Freud (1915: 116-121) considers that
very argument and provides compelling counter-arguments. Searle acknowledges those argu-
ments, but dismisses them summarily.
29. Incidentally, Wittgenstein (1980, 1980b) said that Freud wasn’t producing a substantive
theory, and that he was really only suggesting that we play a new “language game” – that the
rules governing the use of expressions such as “mental state” and “unconscious” be changed.
I believe that Wittgenstein’s point is absurd. For reasons that I am now in the process of
explicating, the question whether Freud’s theories are right is a strictly empirical one, at least in
so far as it is not a matter of logic that is to be decided in Freud’s favor, and is therefore not to any
degree a question of preferring one mode of speech to another.
 Conceptual Atomism and the Computational Theory of Mind

similar to, a surge of joy, a perception, an urge, and so on. Nothing could be rele-
vantly similar to such things without itself being a constituent of consciousness.
Thus, anything that is relevantly like our paradigms of mentality must itself be a
constituent of consciousness. We therefore eviscerate the concept of mind if we
say that there are mental entities that are not constituents of consciousness, just as
we eviscerate the concept of physicality if we say that desks, rocks, and trees are
not physical objects.
Of course, there are dispositional mental entities – e.g. beliefs, aspirations,
propositional attitudes generally, as well as character traits – and these cannot be
identified with constituents of consciousness. (Although it has conscious manifes-
tations, my belief that 1+1=2 is not itself comparable to a tickle or a pain: it doesn’t
vanish when I lose consciousness.) But this isn’t a significant problem for the view
that mentality is identical with consciousness. A belief that 1+1=2 is nothing more
than a disposition to have certain conscious states and to engage in certain behav-
iors, given certain inputs, e.g. a disposition to think Smith must have two apples
upon being told Smith used to have one apple and he has now been given a second
one (and he never ate, or otherwise parted with, the first one). That disposition can
be understood in strictly neurological terms.
In general, so far as beliefs, character traits, and so forth, are distinct from their
conscious manifestations, they are to be understood in strictly physiological terms:
as physiological propensities to have certain conscious states (as well as to engage in
ceratin behaviors). Hearts, livers, and bile-ducts give rise to conscious states. But it
would obviously be absurd to say that such things were components of one’s mind.
It would be similarly absurd to say that, because they give rise to conscious states,
neural structures are veritable mental entities. 30
Anything that is mental is a conscious state. Everything else is pure physiol-
ogy. So-called non-occurrent mental entities are purely physiological dispositions
to have certain conscious mental states.

This view is untenable, the reason being that, as we will now see, it implicitly denies the
existence of beliefs, aspirations, and propositional attitudes generally.
We must begin by making a preliminary distinction. Right now I am conscious of
the moon. (I can see it through the window behind my computer.) But the moon is not
a constituent of my consciousness. Rather, it is an object of my consciousness. I cannot,
at any given time, be aware of every one of my own awarenesses. In other words, those
awarenesses cannot all be objects of my awareness. It is therefore incontrovertible that
some of our own awarenesses are “unconscious” in the sense that we are not, at all
times, conscious of them. So those who identify mentality with consciousness cannot
mean that something is a part of a person’s mind only if that person is conscious of it.
They must mean that something is a part of a person’s mind only if that thing is a con-

30. See Searle (1992). Searle adopts a dispositionalist view of many cognitive abilities. Searle
also rejects functionalism. But dispositionalism is a form of functionalism. So, in this respect,
Searle’s views are internally inconsistent.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

stituent of his consciousness. For argument’s sake, let us suppose that this is correct. In
that case, for the very reasons given a moment ago, it is necessary to take a disposition-
alist view of propositional attitudes. For argument’s sake, let us do this. Let us therefore
suppose that my belief that 1+1=2 is in fact a disposition (to have certain conscious states
and to engage in certain behaviors, given certain sensory inputs). Let B be the brain-state
(or neural-structure) that mediates that belief, i.e. that disposition or set of dispositions.
A story will help us move forward. I am told “Bill used to have exactly one apple; he
didn’t eat it or otherwise lose possession of it; and he has just been given another apple.”
In response I say: “In that case, Bill now has two apples.” I make these noises for the right
reasons: I make them because I know what they mean, and because I know that, under
the circumstances, that meaning is one that can correctly be affirmed.
As we discussed earlier, it cannot be B simpliciter that causes me to produce that ut-
terance. (It is not the rock that breaks the window – it is the rock’s moving with a certain
velocity…) If it isn’t B’s having a certain content that causes me to produce that utterance,
then we don’t have a case where my utterance manifests a belief. Suppose that it was B’s
thermal, not its representational, properties that led to my utterance. (For the sake of
argument, suppose that B’s representational properties are not realized by its thermal
properties.) In that case, my utterance was no more a manifestation of arithmetical
knowledge than an involuntary muscle-spasm. In fact, that utterance would, under those
circumstances, be a mere by-product of just such a spasm. A body-movement is volun-
tary only if the person whose body is involved wants it to happen. To want something is
to have a certain kind of attitude towards a proposition (or, in any case, towards some
kind of content). So a body-movement is voluntary only if it results from someone’s hav-
ing an attitude towards a proposition and, therefore, from something’s being a represen-
tation of something. So if it wasn’t B’s (or, if not B’s, then something else’s) being a repre-
sentation of some kind that led to my utterance, then that utterance could not have been
something that I wanted to happen and thus could not have been voluntary. It would
therefore have resulted from an involuntary muscle-spasm and would no more repre-
sent a belief than would a hiccup. Thus, supposing that my utterance is not comparable
to a hiccup and that it does embody arithmetical knowledge, that utterance must be an
effect of B’s being a belief that 1+1=2 (or, in any case, it must be an effect of something’s
being a representation of something). Thus, to have a belief, it is not enough to have the
right dispositions. One must have the right dispositions for the right reasons. If it isn’t B’s
(or, if not B’s, then something else’s) representational properties that dispose me to say
“Bill now has two apples”, then that utterance is no more an expression of arithmetical
belief than an epileptic fit. But if it is B’s representational properties that dispose me to
make those noises, then that disposition is an effect of my belief, and is therefore not
constitutive of it.31
This isn’t to say that my belief that 1+1=2 is not a physiological entity or, more pre-
cisely, that it is not identical with some physiological entity’s having some physiological

31. Pap makes a similar point. See Pap (1962: Chapter 20), especially pages 393-400.
 Conceptual Atomism and the Computational Theory of Mind

property; it isn’t to say that B’s encoding the relevant information isn’t identical with, or
otherwise supervenient on, B’s having some property that falls within the scope of phys-
ics (or biology). It is to say that if B’s encoding the relevant information is realized by its
having some physical (or biological) property, then some physical (or biological) prop-
erty realizes the psychological property of encoding a certain piece of information.
Now we can close the argument. B’s having those psychological properties cannot
be identified with my being in any conscious state. (There are two reasons for this.
First, B has those properties even while I am unconscious. Second, any conscious state
is, at most, a causal consequence of B’s having those properties and must therefore be
distinct from B’s having those properties. In any case, everybody agrees that my belief
that 1+1=2 isn’t a constituent of consciousness.) Since, as we have seen, B’s having
those properties is not identical with any disposition, it follows that B’s having those
properties is a bona fide unconscious mental entity, i.e. it is a mental entity that is not
a constituent of consciousness. It is in virtue of the fact that B has those properties that
you believe that 1+1=2. Thus, your belief that 1+1=2 is a mental entity that is not a
constituent of consciousness. Consequently, if that belief is “conscious”, it is in the
sense that it is an object of consciousness. But we’ve already seen one reason why it
would be absurd to suppose that one was aware of all of one’s own awarenesses; and we
will encounter other such reasons when, in a moment, we are discussing Sartre’s views
on the unconscious. 32

32. Dispositionalism (the view that our propositional attitudes are merely dispositions to react
in certain ways to certain stimuli) turns us into robots. If my belief that 1+1=2 is just a disposi-
tion to react in certain ways to certain perceptual stimuli, then ipso facto those reactions are not
mediated by any kind of understanding and are therefore unprincipled and, hence, robotic. Dis-
positionalism is, in fact, nothing other than Skinnerian behaviorism. And if we have learned
anything in the last fifty years, it is that behaviorism is wrong. Nonetheless, even though few
philosophers would admit to accepting behaviorism (even though neo-Wittgensteinians such as
John McDowell do accept it, though not by that name), and even though that doctrine strips
anyone of the ability to react in a principled manner to anything, nonetheless most philosophers
accept a dispositionalist analysis of propositional attitudes. (Chomsky is a conspicuous excep-
tion. See Chomsky 1988.) Despite its obviously having false consequences, the dispositionalist
view is extremely popular. But why?
I believe that, ultimately, the popularity of dispositionalism results from a failure to realize
that what we think of as situations that involve no change are in fact situations that involve
constant, as opposed to variable, change. Consider an object (some rock, let us suppose) that
would not ordinarily be described as changing. Every moment of its existence involves innume-
rable displacements of mass-energy. This is not to say that there isn’t a profound difference
between an object that would be described as undergoing change, e.g. a building that is being
torn down, and one that is not. Obviously there is such a difference. But the difference is not that
the building does, whereas the rock does not, instantiate various changes. The difference is that,
in the case of the rock, the changes are structure-internal, whereas in the case of the building
they are structure-external. Although the rock is characterized by various changes, those chan-
ges do not themselves change. The changes that occur are regular. This is not the case with the
 Chapter 18.  Fodor’s third argument for conceptual atomism 

building. Here we are dealing, not merely with changes, but with changes in the manner in
which changes occur. Supposing that the building is destroyed at time t, the patterns of energy-
exchange that constituted the building before t have either been altered or obliterated.
A physical object is no more changeless than a nation. A politically and economically stable
nation is teeming with economic and political activity. Such a nation is not one where there is no
political or economic activity or change: it is one where such change happens in a constant man-
ner. There is constant change; but those changes are internal to structures that do not themselves
change, and there is thus no change in the manner of change. But if there is a coup d’êtat, and a
consequent change of government, there is a change in the manner in which those changes oc-
cur. As we’ve seen, an analogous point holds of any large or medium-sized object.
We must distinguish conditions from events. An event is a change in a manner of change: a
hyper-change. A condition is a situation of constant change. The difference between events and
conditions is not to be understood in strictly chronological terms. An event can take years; a
condition can last for a day. World War II took six years. But a poorly built building that col-
lapsed a nanosecond after being erected would, for that period, constitute a condition. In a
moment we will see the relevance of this point to our inquiry.
To close this line of thought, I am going to make some highly conjectural points. It seems
to me that psychological entities rise above the threshold of consciousness only if they are events.
(But, of course, there are innumerable psychological events that remain well below that thres-
hold. We are dealing, at most, with necessary, not sufficient, conditions for becoming conscious.)
Consciousness seems to track disruptions. One ceases to be consciously aware of a noise that
doesn’t change. (This is, of course, only approximately true. For example, it doesn’t hold of extre-
mely loud noises.) One becomes aware of a background hum when there is some kind of change
– when the hum stops, or changes pitch or timbre, or becomes louder. But notice that this chan-
ge is really a hyper-change. The hum itself was a change: it represented a continuous jet of audi-
tory stimulation. But this change did not itself change: the jet didn’t become more or less in-
tense. But when this change itself changed – when it ceased to exist or doubled in loudness – then
the subject suddenly became aware of it.
One could explain this by saying that there is now a perception where previously there was
no perception prior to the change (that is, to the hyper-change). But that view strikes me as ra-
ther arbitrary. It posits an ex nihilo change, thus creating a causal anomaly (where there is no
reason, apart from what we have seen to be a dubious antagonism towards the unconscious, to
believe that anything anomalous is occurring). It also conflicts with an intuition that, all along,
the perception in question lay just outside the periphery of consciousness.
I thus follow Leibniz in saying that the perception was indeed present all along. But now,
because of the change in the hum’s loudness, instead of merely having a perception of the hum,
the subject pays attention to that perception; and it is because it is now the object of scrutiny that
the perception has become a constituent of consciousness. The change consists in the subject’s
focusing on some perception that he had all along, but previously paid no attention to. Freud
(1915: 40) put it by saying that the perception has become “hyper-cathected”, this being why it is
now conscious. Freud’s view is that the contents of consciousness are categorically hyper-cathec-
ted. It seems that recent so-called “higher-order thought” (HOT) theories of consciousness re-
present a kind of echo, or recrudescence, of Freud’s view. See, for example Rosenthal (1986).
If this line of thought is correct, then strictly speaking what is conscious is not the percep-
tion, but one’s awareness thereof. (This was Freud’s (1915) position.) This shows to what a large
 Conceptual Atomism and the Computational Theory of Mind

extent our theories of consciousness may involve a failure to distinguish the constituents of
consciousness from the objects of consciousness. What we assume to be constituents of
consciousness turn out to be objects of consciousness. When the perception “becomes conscious”,
what is really happening is that there comes into existence conscious awareness of it. The percep-
tion per se doesn’t necessarily change. What is a veritable constituent of consciousness is not the
perception itself, but is rather one’s awareness of it. (Again, this was Freud’s (1915) view.) If that
is right, then antipathy towards the notion of unconscious mentation expresses a failure to dis-
tinguish between the notion of an object of consciousness and an object’s being a constituent of
consciousness.
In any case, where there is psychological changelessness (or, more accurately, where there is
constancy of psychological change) there is a tendency to become unconscious; and where there
is psychological change (or, more accurately, change in the manner of change), there is a ten-
dency to become conscious. These facts suggest that consciousness is associated, not with psy-
chological existence in general, but rather with changes in psychological conditions – or, indeed,
with changes in the manner in which psychological conditions change.
Admittedly, this argument must be qualified. While there is clearly some tendency for un-
conscious changes (or hyper-changes) to enter consciousness, this tendency is a weak one. The-
re is no tendency for subpersonal hyper-changes to become conscious. And the internal disequi-
libria that, according to Freud’s theories, constitute the bulk of our psychological lives remain
below the threshold of consciousness.
But we must not be too quick to take these points to warrant a rejection of our conjecture
that consciousness is to be understood in terms of psychological hyper-changes and uncons-
ciousness in terms of an absence of such changes. If Freud’s theory, or anything even remotely
like it, is correct, it is only because of special resistances that those disequilibria fail to come to
the attention of consciousness. Also, their symptomatic derivatives do become conscious. (Othe-
rwise we would know nothing of them, as Freud himself emphasized.) We might say that they
do become conscious, albeit in pathologically distorted forms. It thus seems that repressed psy-
chological events form no counterexamples to our principle that ceteris paribus consciousness
consists of hyper-changes (or, more accurately, of our consciousness of such changes) – just as ,
to use Mill’s example, the floating of a helium-filled balloon is no counterexample to the princi-
ple that gravity tends to draw objects nearer to one another.
Of course, there is no tendency for subpersonal changes (or hyper-changes) to gain entry
into consciousness. But I don’t think that this demands a rejection of our principle. So far as one
has a tendency to think otherwise, one is guilty of a confusion similar to that discussed a mo-
ment ago.
A metaphysical digression is needed to develop this analysis. Some events happen within
frameworks without themselves being constitutive of frameworks; and other events are themsel-
ves constitutive of frameworks. The activities of police-officers, judges, and administrators
constitute social frameworks within which people operate. When I stop at a stop-sign, I am
operating within that framework; but my behavior is not constitutive of it. My behavior is a
causal effect of that framework and therefore cannot be a part of it.
This line of thought generalizes. A building is made of bricks; those bricks are composed of
mass-energy displacements. Those displacements are thus constitutive of a certain structure or
framework. A mouse that is running through the heating vent of that building is operating wi-
thin that framework, but its activity is not constitutive of it. The impact of a wrecking-ball with
 Chapter 18.  Fodor’s third argument for conceptual atomism 

the side of the building is not only not constitutive of the structure of the building, but also fails
to occur within that structure: in fact, it damages that structure. The activity of somebody
playing basketball ten miles from the building is not constitutive of the structure of the building;
nor does it happen within that structure; nor does it alter that structure.
In light of these points, let us resume our discussion of the subpersonal. The subpersonal
events posited by Chomsky and Marr constitute our cognitive framework. The conditions and
events associated with the personal level of mentation presuppose subpersonal activity. Sense-
perceptions occur at the personal level of mentation. But sense-perceptions are the outcome of
innumerable subpersonal processes. These processes never gain entry into consciousness, or
into any other facet of the personal level of mental activity, because they provide the framework
within which conscious and, more generally, personal activity occurs. The mass-energy displace-
ments that compose a building do not become events that happen within the building – unless
some kind of structural breakdown occurs.
Events that are constitutive of a structure do not occur within that structure. Because sub-
personal cognition forms the structure that houses personal psychological activity – the activity
consisting of perceiving, wishing, intending, believing, and so forth – such cognition cannot it-
self enter the sphere of personal cognitive activity. The fact that the subpersonal cannot enter the
personal cognitive sphere is thus not a significant challenge to our principle that hyper-changes
tend to become conscious, whereas constant changes do not. That fact is simply an expression of
the general metaphysical principle that the events that that compose a structure do not happen
within that structure, since they are that structure.
I would like to develop the points made a moment ago concerning the difference between
structure-internal and structure-external changes. Spatiotemporal reality doesn’t consist of scat-
tered events. Events often happen within structures. If they did not, then reality would be far less
amenable to explanation and prediction than it actually is. Because I know that Jim works for a
corporation, and that he has certain duties, I can predict what he will do with a fairly high-de-
gree of accuracy. If Jim operated outside of any structure – if he didn’t have a job in a corporation
and weren’t in any other comparable situation – then it would become very difficult to predict
his behavior: I would need special knowledge about his internal processes. But in so far as Jim is
merely a part of a system, such high-resolution, Jim-specific knowledge isn’t necessary to know,
in broad terms, what he will do on a given day. (This point is powerfully defended in Jackson and
Pettit 2004d.)
Another example: Because I know that brick R is part of a stable building, I have a pretty
good idea where it is going to be tomorrow. But if that brick were lying on a side-walk, I would
have a much harder time making predictions as to its whereabouts. To make such a prediction,
I would need quite specific knowledge about the brick’s environment – about the people who
live there, their attitudes towards bricks that are lying on the side-walk, the meteorological and
seismic conditions that obtain in that area, and so on.
Where there is structure, there is constancy in change, and accurate predictions can be made
on the basis of highly generic knowledge. Where there is no structure, predictions can be made
only on the basis of highly specific knowledge (Jackson and Pettit 2004d). (This, incidentally, is
one of the reasons why higher-level sciences (e.g. economics) are not merely inferior versions of
physics. Given the data available to an economist, a microphysicist couldn’t even begin to tell you
why the stock-market crashed. But given that same data, the economist has no trouble doing so.
Higher-level sciences identify principled connections in data in which lower-level, more funda-
 Conceptual Atomism and the Computational Theory of Mind

mental sciences fail to find principled connections. So supposing that it is the objective of scien-
ces to organize data, there are things that economics can do that microphysics cannot do, and it
is therefore wrong to say that economics is simply a bad version of microphysics. It may be that
economics is, in fact, a bad science. (This is not my view, by the way.) But it wouldn’t be bad on
account of not being microphysics; it would be be bad because it couldn’t identify principled re-
lations in the data that falls within its bailiwick. I think that Carnap’s (1934) view that all sciences
are either microphysics or are, at best, bad versions of microphysics stems from a failure to see
that sciences model data and not, except indirectly, the things that generate that data. (We tou-
ched on this when we discussed Davidson’s views on self-knowledge.) Even though the things
studied by biology are among the things studied by physics, the data that the former attempts to
model is different from the data that the latter attempts to model. Since sciences aspire to find
principled concomitances in the data that they are concerned with, and not in the underlying
things (except by way of finding principled concomitances in the relevant bodies of data), it fol-
lows that there is no reason why biology ought to be assimilable to physics.)
We tend to think of those structures as not involving changes of any kind. But that is false:
those structures consist of changes. The building consists of mass-energy displacements. Struc-
tures consist of constant changes – changes that do not change. Events consist in changes in the
manner of change – they consist of hyper-changes. Given a structure, a given hyper-change may
happen within that structure, while not being constitutive of it, or it may not happen within the
structure. If an event doesn’t happen within a structure, and it isn’t constitutive of that structure,
it may or may not alter that structure.
We have a tendency to think of frameworks as inert – as involving no change. But this is
false, as we’ve seen. Frameworks consist of constant change – in both senses of the word
“constant.” Those changes are incessant (so they are “constant” in that sense). But those changes
do not themselves change (so they are “constant” in that sense). What we think of as events are
occurrences that are not constitutive of structures. Where there are frameworks, there is predic-
tability and a consequent absence of disruptions. For obvious reasons, we focus on disruptions,
not on constancies. For this reason, we come to see structures, like buildings, as involving no
change. We tend to think of them in strictly dispositional terms.
Given that conception of what a structure is, it is a short-step to saying that structures just are
the regularities that are associated with them. Everything that is not an event becomes a mere
disposition to create events. So everything that is not an event becomes nothing more than a ten-
dency for events to occur with a certain regularity. This means, ultimately, that only events exist.
But this view involves a failure to make a number of distinctions: the distinction between
changes (conditions) and hyper-changes; the distinction between an occurrence that is constitu-
tive of a structure and an occurrence that takes place within a structure; and, finally, the distinc-
tion between an event that happens within a given structure, while not constituting that struc-
ture, and an event that does not happen within a structure.
Given these distinctions, there is little room in reality for “mere dispositions.” Once the
notion of a mere disposition is gone, so is the idea that beliefs, character traits, and “non-occur-
rent” mental entities generally could be mere dispositions. The idea that a belief is nothing other
than a disposition for certain inputs to bring about certain outputs turns out to be a special case
of the fallacious view that, where there are no events, there is nothing at all (except possibly for
some kind of indefinable “potentiality”). The latter view, in its turn, is an expression of the failu-
re to realize that what we refer to as “events” are hyper-changes, as opposed to changes simplici-
 Chapter 18.  Fodor’s third argument for conceptual atomism 

Sartre on repression and the unconscious

Like Searle, Jean-Paul Sartre (1943) argued that the very idea of unconscious thought
is an absurd one. The fallacies in Sartre’s argument are to be seen in terms of develop-
ments of points made in the last section. (Some repetition of those points will be nec-
essary.) Here is my reconstruction of Sartre’s argument:
The term “unconscious consciousness” is a contradiction in terms. Beliefs, desires,
and propositional attitudes generally are forms of consciousness. After all, they
are awarenesses, and awarenesses are consciousnesses – to be aware of something
is to be conscious of it. (Of course, propositional attitudes are not necessarily
awarenesses of facts. Your intention to become a great lawyer is not identical with
an awareness of some fact. But a propositional attitude necessarily involves an
awareness of a proposition. Your intention to become a great lawyer involves an
awareness on your part of the proposition that you will be a great lawyer.) Thus,
the very concept of an unconscious belief (or desire or intention…) is an incoher-
ent one, since it is nothing other than the idea of an unconscious consciousness
and thus of a consciousness that is not a consciousness or, equivalently, of an
awareness that is not an awareness.

To identify the fallacy in this argument, we must recall the distinction, discussed in the
previous section, between a thing’s being an object of consciousness, on the one hand,
and its being a constituent of consciousness, on the other. Some mental entities are
constituents of consciousness, i.e. they are themselves consciousnesses. An itch is a
consciousness. (Of course, it may also be an object of consciousness. I can think about
an itch that I am having.) The same is true of all sensations. But propositional attitudes

ter, and that what we refer to as “stable conditions” are constant changes and are therefore cau-
sally pregnant.
Hume argued that a causal sequence is a nothing other than a juxtaposition of events that
instantiates some kind of regularity. This conception of causality involves the view that there
isn’t really anything where there are no events. It involves the view that there are changes only
where there are events and that, consequently, where there are no events, there is nothing that
can do anything. Anything that is not an event becomes, at most, a “disposition” for events to
occur, and thus becomes nothing.
We have seen at least one reason why such a view must be questioned. There is constant
change where there are no events. There is thus causality where there are no events, given that
change involves causality. Because it applies only to events – i.e. only to changes in the manner of
change, and not to change in general – Hume’s analysis of causality is severely limited in scope.
In fact, given that hyper-changes are presumably to be understood in terms of changes, and
given that Hume’s analysis is silent as regards the nature of the causal forces governing the latter,
it becomes questionable whether Hume’s analysis has anything viable to say about any kind of
causation. My guess is that, once we register the various distinctions that we’ve been discussing,
it becomes impossible even to articulate Hume’s analysis of causality. But that is a matter for
another work.
 Conceptual Atomism and the Computational Theory of Mind

are not constituents of consciousness. Your intention to become the world’s greatest
defense-attorney is not a constituent of consciousness. Nor, as we saw earlier, is it a
disposition. No propositional attitude is a constituent of consciousness and no propo-
sitional attitude is a disposition. For a propositional attitude to be “conscious” is for it
to be an object of consciousness.
If x and y are distinct, then x can exist in the absence of y. (Even though I wouldn’t
have come into existence if it weren’t for my parents, I am now capable of existing in
their absence. So even though my existence is constitutively dependent on certain facts
about their prior conduct, and is therefore constitutively dependent on their having
existed, I can now exist in their absence. And the reason, of course, is that I am distinct
from them.) Our propositional attitudes are not constituents of consciousness. Any
propositional attitude is distinct from any consciousness, and can therefore exist inde-
pendently of that consciousness. (And, for reasons that parallel those made in the pre-
vious parenthetical remark, this holds even if, for some reason, it should turn out that
consciousness is needed to initiate propositional attitudes.) It follows that there needn’t
be any consciousness of any propositional attitude. Thus our propositional attitudes do
not depend for their existence on our being conscious of them. Strictly speaking, prop-
ositional attitudes are never conscious; that is, they are never conscious in the sense in
which sensations are conscious. (See Freud 1915: 116–120.) They are conscious only in
the sense that we may be conscious of them. (We are conscious of tables and chairs.
And yet we would never dream of describing such things as “conscious.” This shows
how muddled commonsense views about consciousness are.)
For a propositional attitude to be unconscious is for one not to be conscious of it.
It is not for one to be unconsciously conscious of it. If my belief that my friend is evil is
unconscious, then I simply don’t know about it. I don’t know about it consciously or
unconsciously. My belief that my friend is evil is itself an awareness. (To echo a point
made a moment ago, it isn’t necessarily an awareness of a fact, but it is necessarily an
awareness of a proposition.) But if that attitude is unconscious, nothing is conscious of
it; nothing knows about it consciously or unconsciously. Sartre thinks that, were such
a thing to exist, an unconscious thought would be a thought of which one were uncon-
sciously conscious. Of course, if that were the case, then the concept of an unconscious
thought would be incoherent. But an unconscious thought is not one of which one is
unconsciously conscious. It is simply one of which one isn’t conscious.
Of course, any belief, whether conscious or not, is itself an awareness (of a propo-
sition). But it doesn’t follow that, in order for it to exist, there must be some awareness
(or consciousness) of it. An awareness doesn’t need to be the object of a second aware-
ness in order to exist. (Sartre denies this, holding that, in fact, a precondition for the
existence of any awareness is that there be some awareness of it. We will soon discuss
Sartre’s reasons for holding this obviously incoherent view.)
The sentence:
(1) “there are unconscious consciousnesses”
 Chapter 18.  Fodor’s third argument for conceptual atomism 

is ambiguous, as it could mean either:

(2) “there are awarenesses that are not awarenesses”
or
(3) “there are awarenesses that are not themselves the objects of awareness.”
(2) is tautologously false, and (2) is the meaning that Sartre assigns to (1). But (3), not
(2), is the correct disambiguation of (1). Moreover, (3) is true. In fact, given that there
are instances of propositional attitudes, (3) is analytically true, since it would be inco-
herent, because viciously regressive, to suppose that one were aware of every one of
one’s own awarenesses.
To say that we have “unconscious” beliefs and desires is to say that we have beliefs
and desires that we are not aware of. It is not to say that we have beliefs and desires that
are not themselves awarenesses of anything. Freud didn’t hold that one could have
propositional attitudes that weren’t awarenesses of anything. He held that proposi-
tional attitudes that one is not aware of are among the constituents of one’s psyche.
Sartre’s argument appears probative only so long as we fail to recognize that “con-
scious” is ambiguous in the way just described.
Sartre’s argument involves a failure to register another, similar ambiguity. The term
“consciousness” sometimes refers to phenomenologically pregnant awarenesses, and it
sometimes refers to awarenesses that don’t have that property.33 Your belief that 1+1=2
is an awareness: it is an awareness of a proposition and also, because that proposition
is a true one, of a fact. But, as we’ve discussed, that awareness is not, and couldn’t pos-
sibly be, a constituent of consciousness; and it therefore is not, and couldn’t possibly be,
“conscious” in the sense of being phenomenologically pregnant. If we use the term
“conscious” to refer to phenomenologically pregnant states, then your belief that 1+1=2
is not conscious and is thus an “unconscious awareness.”
Thus, we see that, contrary to what Sartre alleges, the expression “unconscious
awareness” is not an oxymoron. In the expression “unconscious awareness”, the word
“unconscious” means “lacking phenomenology”, not “lacking awareness.” Since there
is nothing absurd in supposing that there are awarenesses that lack phenomenology,
the term “unconscious awareness” is not an oxymoron. In fact, given that some aware-
nesses are propositional attitudes, it would actually be absurd to suppose that all aware-
nesses did have phenomenology. (Of course, not all awarenesses can be described as
“unconscious”, given that sense-perceptions are awarenesses that are obviously con-
stituents of consciousness.)
Let us now close the argument. Freud’s position that there are unconscious aware-
nesses is often identified with the position that there are unconscious consciousnesses.

33. Incidentally, if there is subpersonal thought, then not all occurrent processes are phenome-
nologically pregnant. If Chomsky is right, language-comprehension is mediated by occurrent
processes that, being subpersonal, don’t have any phenomenology.
 Conceptual Atomism and the Computational Theory of Mind

(Sartre explicitly makes this identification.) It is then argued that, since the term “un-
conscious consciousness” is obviously a contradiction in terms, the very idea of uncon-
scious ideation is an incoherent one. But, as we’ve just seen, “unconscious conscious-
ness” is ambiguous and, for reasons that have nothing to do with psychoanalysis, that
expression, appropriately disambiguated, picks out an extensive class of mental states.
Thus, the charge that “unconsciousness consciousness” is a contradiction in terms can-
not be sustained.
Incidentally, Freud (1915: 127) explicitly says that phenomenology cannot be un-
conscious and that it is only propositional attitudes that can have that property. (Freud’s
argument for this is obscure. But, so far as I can make it out, it appears to coincide with
ours.) Freud himself holds that there cannot be unconscious “consciousnesses” if that
term is taken to mean “phenomenologically pregnant awarenesses.” The essence of
Freud’s view is that what is not phenomenologically pregnant needn’t always be an
object of consciousness. And we have seen that, although it is obvious that some prop-
ositional attitudes are objects of consciousness, there is no a priori reason why some
should not have that property.
In fact, as Freud (1915) himself points out, it is obvious that many propositional
attitudes are unconscious: surely you are not, at this very moment, conscious of all of
your beliefs. (Freud describes those propositional attitudes that one knows oneself to
have as “preconscious” or “topographically unconscious.” And he describes those
propositional attitudes that one has, but does not know oneself to have, as “dynami-
cally unconscious.” Everything that is repressed is dynamically unconscious. Actually,
“repressed” and “dynamically unconscious” are synonyms. Later, when we are analyz-
ing Sartre’s critique of the concept of repression, we will further discuss the difference
between topographical and dynamic unconsciousness.)
As we parenthetically noted a little while ago, Sartre (1943: 13) holds that there is
no difference between being a person’s being conscious of X and his being conscious
that he is conscious of X (the italics are Sartre’s own):
[I]n order to count, it is necessary to be conscious of counting…to know is to
know that one knows. Of course, someone may say, but this makes a circle. For is
it not necessary that I count in fact in order to be conscious of counting? That is
true. However, there is no circle, or if you like, it is the very nature of conscious-
ness to exist in a circle. The idea can be expressed in these terms: Every conscious
existence exists as consciousness of existing.

Of course, if Sartre is right, it immediately follows that there is nothing in a person’s

mind that he is not aware of. The problem is that, despite what Sartre says, his position
clearly is viciously circular. Sartre’s statement that “it is the very nature of consciousness
to exist in a circle” does not ameliorate the situation. This is because, for reasons of a
generally logical nature – reasons that have nothing to do with consciousness specifi-
cally – nothing (including consciousness) can exist “in a circle.” Meaningful sentences
cannot refer to themselves. An utterance of the sentence “this sentence – the one I am
 Chapter 18.  Fodor’s third argument for conceptual atomism 

now uttering – is false” is neither true nor false, and thus meaningless, the reason being
that it refers to itself. Sartre’s statement that “consciousness is self-consciousness” is
characterized by the same kind of circularity; and in attempting to exonerate his state-
ment by saying that “it is the nature of consciousness to exist in a circle”, Sartre is asking
that the laws of logic be suspended where his analysis of consciousness is concerned.
Aware that few will accept his statement that “it is the nature of consciousness to
exist in a circle”, Sartre makes another attempt to show that his analysis involves no vi-
cious circularity. This attempt involves a distinction between two different kinds of self-
consciousness. On the one hand, there is “positional self-consciousness.” This is the
kind that I have if I think to myself: I am now thinking that there are twelve cigarettes on
the table. Positional self-consciousness is what would ordinarily be described as “self-
consciousness.” On the other hand, there is “non-positional” self-consciousness. This is
the kind that I have when I think there are twelve cigarettes on the table. Non-positional
self-consciousness is what would ordinarily not be described as “self-consciousness.”
Sartre says that, although it would indeed be viciously regressive to suppose that (a) all
consciousness was positionally self-conscious, it is not regressive to suppose that (b) all
consciousness is non-positionally self-conscious.
Here two closely related questions arise. First, why is (b) not viciously regressive?
Second, what is the difference between a “non-positional self-consciousness”, on the
one hand, and an awareness that is not the object of any awareness, on the other? In
other words, why say that so-called non-positional self-consciousnesses are in fact
self-consciousnesses? Why not just say that they are consciousnesses that are not self-
consciousnesses? After all, it seems straightforwardly false to say that I am being self-
conscious in believing that there are twelve cigarettes on the table.
Sartre (1943: 13) gives an answer to the second question and thus, arguably, gives
an implicit answer to the first (the italics are Sartre’s own):
If I count the cigarettes which are in that case, I have the impression of disclosing
an objective property of this collection: they are a dozen. This property appears to
my consciousness as a property existing in the world. It is very possible that I have
no positional consciousness of counting them. Then I do not know myself as
counting. Proof of this is that children who are capable of making an addition
spontaneously cannot explain subsequently how they set about it…Yet at the mo-
ment when these cigarettes are revealed to me as a dozen, I have a [non-position-
al] consciousness of my adding activity. If anyone questioned me, indeed, if any-
one should ask, “What are you doing there?” I should reply at once, “I am
counting.” This reply aims not only at the instantaneous consciousness which I
can achieve by reflection but at those fleeting consciousnesses which have past
without being reflected on, those which are forever not-reflected-on in my im-
mediate past. Thus reflection has no kind of primacy over the consciousness re-
flected-on. It is not reflection which reveals the consciousness reflected-on to it-
self. Quite the contrary, it is the non-reflective consciousness which renders the
 Conceptual Atomism and the Computational Theory of Mind

reflection possible; there is a pre-reflective [non-positional] cogito which is the

conditions of the Cartesian [positional] cogito.

Here is my interpretation of what Sartre is saying:

In counting the cigarettes, I am necessarily conscious of counting the cigarettes.
How do we know this? First of all, if you ask somebody who is counting cigarettes
what they are doing, they will tell you without a moment’s hesitation. They don’t
have to think, and this shows that their knowledge of what they are thinking is
built right into their thinking of it. If my counting the cigarettes were not moni-
tored – if there weren’t some internal awareness of it – then it would go unre-
corded and, immediately upon completing it, I would have no recollection of it.
Given any experience of mine, an exactly analogous argument shows that, unless
it were accompanied by some awareness of it, I couldn’t possibly have any recol-
lection of it. Thus, every experience of mine that can be remembered must have an
awareness of itself built into it. Since every experience that one has could (in prin-
ciple) be remembered (even though many of them are forgotten), it follows that all
consciousnesses are really self-consciousnesses.

There are obviously some cases – indeed, a great many cases – where people know ex-
actly what it is that they are doing. Sartre is making a plausible point in saying that, if
somebody is counting cigarettes, he doesn’t need to reflect to know that he is doing so.
But there are two reasons why this point doesn’t have the significance that Sartre as-
cribes to it.
It is true that, in creatures (such as human beings) that have sophisticated and high-
ly stratified minds, one is often conscious of performing a task as one is performing it.
But this doesn’t mean that a necessary component of one’s performing a task is one’s be-
ing conscious of that same performance. When a pianist sight-reads a piece of music, he
is deliberately doing many things that he is not aware of doing. And even if he is aware
of them, he doesn’t have to be: it would be absurd, because it would be circular, to say
that one had to be aware of a task in order to perform that task. We are indeed conscious
of much of what we do. But that isn’t because of anything inherent in the concept of de-
liberate action: it is only a reflection of the contingent fact that we are sophisticated
creatures that have highly stratified minds. Do all animals, or even all people, always
know just what it is they are doing deliberately? Surely not. What we just said about the
performing of tasks is true of the having of thoughts. Many of our thoughts are indeed
accompanied by awarenesses of those same thoughts. But it doesn’t follow that thoughts
must be accompanied by such awarenesses. A fortiori it doesn’t follow that thoughts
must have awarenesses of themselves built into them. Sartre appears to have conflated
the truth that thoughts are often accompanied by awarenesses of them with the absurd-
ity that thoughts always have awarenesses of themselves as constituents.
Second, Sartre’s example is an extremely tendentious one and doesn’t support the
generalization that he makes on the basis of it. Suppose that Smith, a highly gifted
musician, prefers composition X to composition Y. It might take Smith years to put his
 Chapter 18.  Fodor’s third argument for conceptual atomism 

finger on just what it is about X that makes him prefer to it to Y. In fact, it is extremely
unlikely that Smith will ever produce anything in the way of an answer to the question
why he prefers X that is more than a hollow platitude (e.g. “X has a certain authentic-
ity, a certain flow, that Y lacks, despite its formal superiority to X”). If Smith is a musi-
cologist, he might, after years of study, begin to articulate his reasons for preferring X
to Y. But even then his answers would be tentative hypotheses. Smith’s situation obvi-
ously isn’t comparable to, and cannot be understood in terms of, that of the cigarette-
counter. The same is obviously true of many other situations, and Sartre is guilty of a
major fallacy in assuming otherwise.
In fact, if you ask someone to provide any of the grounds for any of the judgments
that they make, their answer will almost always be wrong to the extent that it isn’t sim-
ply vacuous. Even though both Smith and Jones are smiling, you know intuitively that
Jones is a malicious liar, whereas Smith is sincere. How do you know this? Your answer
is based on subtle differences in their facial expressions. But unless you have been
trained as a psychological profiler, it would take you hours, even years, to figure out
exactly what the grounds of your judgment were. To take another example: people can
immediately tell whether an utterance of their native tongue is syntactically well-
formed, but only a few savants can even begin to identify the knowledge that underlies
such flash-judgments.
Comparable points can be made about Sartre’s cigarette-counter himself. True – he
doesn’t have to reflect to know that he is counting cigarettes. But he would have to reflect,
and reflect very hard indeed, to know on what exact basis he judged something to be a
cigarette. A comparison might help. Even though it is obvious that “snow is white and
grass is green” is syntactically well-formed, it is not obvious why it is so obvious. Simi-
larly, even though it is obvious whether something is a cigarette, it is not obvious what
makes it so obvious. This suggests that pedestrian judgments (“that’s a cigarette” “that’s
a grammatical sentence”) often have psychological underpinnings that are not immedi-
ately available to conscious introspection. And this in turn suggests, if doesn’t outright
entail, that there is unconscious ideation.
In defense of Sartre’s position, one might say that our flash-judgments about mu-
sic, language, psychology, and so on, are unprincipled or, in any case, do not result
from a knowledge of the relevant principles. It may be that some fact about Jones’ facial
expression causes me to believe that he is a liar, and it may be that all people with that
facial expression are liars. But, according to the view in question, it doesn’t follow that
I knew that fact. It doesn’t follow that I knew the relevant principles. It could be that,
for whatever reason, I am so constituted that other people’s facial expressions simply
cause me to have the right views as to the whether they are being truthful. Comparable
positions could be taken with respect to our linguistic and esthetic judgments. (This is
the position that behaviorists and Wittgensteinians take.)
Is such a view sustainable? If that position is correct, then a person’s ability to under-
stand sentences that he hasn’t previously heard would be identical with a mere disposi-
tion to react in certain ways to certain noises and ink-marks – a disposition that, by
 Conceptual Atomism and the Computational Theory of Mind

hypothesis, was not grounded in a grasp of the relevant principles. Thus, your disposi-
tion to assign the right meaning to an utterance of “snow is white” would not be an effect
of your having the relevant linguistic knowledge. Rather, that disposition would be your
linguistic knowledge (i.e. it would be a part of it). But somebody who doesn’t know that
“snow” means snow, that “white” means white, and so on, doesn’t really understand the
English sentence “snow is white.” Somebody who doesn’t speak a word of English could
be told that “snow is white” means snow is white. But because such a person cannot parse
that sentence – because he doesn’t know how its meaning depends on the meanings of
its constituent-expressions – he still doesn’t understand it and, at most, knows what is
meant by the unarticulated stream of sound “snowiswhite.” Understanding a sentence
involves associating it with the right proposition by way of the relevant rules, and this
condition is not met by someone whose reactions aren’t mediated by syntactic and se-
mantic knowledge. Thus, so far as our linguistic judgments don’t follow from a grasp of
principles, they don’t realize cases of linguistic understanding. Since people obviously do
understand what they hear, it is not an option to say that our linguistic flash-judgments
are categorically unprincipled. Since the relevant principles are obviously not transpar-
ent to one’s consciousness, we must conclude that there is unconscious ideation. Similar
arguments show that our psychological and aesthetic (and spatial and mathematical…)
judgments result from an unconscious awareness of principles.
Another possible defense of Sartre’s position would be to say that the knowledge
that mediates our linguistic judgments is transparent to consciousness, the same being
true of the knowledge that mediates our psychological and aesthetic (and spatial..)
judgments, but that we simply have trouble articulating that knowledge. The idea would
be that everybody who speaks English consciously knows exactly why they associate
snow is white with “snow is white”, but most people have trouble enunciating this knowl-
edge – just as most people have trouble painting what they see. (In some parts of the
Philosophical Investigations, Wittgenstein appears to advocate exactly this view.)
It is not hard to show that this position is false. The debates that rage within the
discipline of semantics concern matters of fact, not matters of nomenclature. The de-
bate between Russellians and non-Russellians doesn’t concern whether we ought to
stipulate that definite descriptions are to be described as “quantifiers”; it concerns the
identity of the rules that people actually use to assign meaning to expressions such as
“the dog in the corner.” Suppose it were to be definitively shown that, whenever an
English speaker hears a sentence of the form F...the phi...G, the rule through which that
person interpreted that sentence was:
(S) If something O is a unique, contextually salient instance of phi, then a token in
that context of F...the phi...G means:…O…

In that case, the Theory of Descriptions would be wrong.

 Chapter 18.  Fodor’s third argument for conceptual atomism 

By the same token, that theory would be proven correct if it were shown that Eng-
lish speakers assigned meaning to utterances of F...the phi...G on the basis of the follow-
ing rule:
(R) If T is a token of F...the phi...G, then T means: something O is a unique, contextu-
ally salient instance of phi and…O…

The question whether (R) is the rule that people actually use is not immediately obvi-
ous to introspection, and the debate between Russellians and non-Russellians isn’t a
debate as to how to label some piece of knowledge that everybody agrees that every-
body has. In general, debates in semantics are not debates about nomenclature; they
are debates about psychological facts whose nature is not transparent to conscious-
ness. (Some authors, e.g. Soames 2001, hold that semantic rules don’t have any psycho-
logical reality. But, for reasons that we considered a moment ago, if that position were
the right one, then nobody would understand any utterance.)
We may conclude that Sartre is wrong to hold that the contents of a person’s own
mind are categorically transparent to his consciousness. Although Sartre is obviously
right that most people who are counting cigarettes are aware that they are doing so, he is
wrong to hold that the phenomenon of self-knowledge can be generally understood in
terms of this one kind of case.
Sartre holds that, unless every consciousness were a consciousness of itself, noth-
ing would be remembered. As we may recall, the idea behind this claim is that con-
sciousnesses must be monitored or kept track of if they are to be stored in memory and
that, since any experience could in principle be remembered, it must therefore be sup-
posed that consciousnesses are all self-monitoring.
The fallacy in this argument is obvious. Given only that something must monitor or
keep track of an experience if that experience is to be remembered, it doesn’t follow that
the experience itself must do so. And it would be absurd, because it would be viciously
circular, to suppose otherwise. We’ve already seen that Sartre fails in his attempts to
show that there would be no vicious circularity in that supposition. Also, it isn’t clear to
me that experiences would in fact have to be monitored to be recorded. Arguably, an
experience could, without the intervention of some monitor, leave some kind of trace
of itself on the basis of which the relevant person’s mind could later have thoughts about
that experience, i.e. on the basis of which that person could remember it.
Sartre’s implausible views about consciousness are direct consequences of his failing
to make two distinctions: the distinction between consciousness and awareness, and the
distinction between being an object of consciousness and being a constituent of con-
sciousness. It is not possible to recognize those distinctions so long as one holds onto the
widely held view that the mind must be transparent to itself. (Let us refer to that view as
the “transparency thesis.”) We’ve seen that the transparency thesis implodes the moment
it is recognized that propositional attitudes are not constituents of consciousness and
that, when they are “conscious”, it is in the sense that they are objects of consciousness.
The transparency thesis thus demands that what are in fact objects of consciousness
 Conceptual Atomism and the Computational Theory of Mind

(namely, propositional attitudes) be absorbed into consciousness itself, a consequence

being that propositional attitudes become inseparable from consciousnesses of those
same attitudes. Thus, Sartre’s absurd view that consciousness is self-consciousness is a
logical consequence of the transparency thesis. We’ve also seen that, once awareness is
distinguished from consciousness, there ceases to be anything paradoxical about the
term “unconscious awareness.” Thus, Sartre’s false assumption that all awareness is con-
sciousness is yet another consequence of the transparency thesis.
Let us now discuss the concept of repression. All repressed thoughts are uncon-
scious (but not vice versa). Therefore, if the concept of unconscious thought is incoher-
ent, then so is the concept of repression. Sartre sees this, and thus regards the concept
of repression as incoherent. But Sartre regards the concept of repression as especially
incoherent, believing it to be guilty of incoherencies additional to those that, in his
view, characterize the concept of unconscious ideation in general. Here is my recon-
struction of Sartre’s (1943: 90–96) argument:
One must obviously be conscious of a thought to initiate the repression of it. But
one must also be conscious of a thought to sustain the repression of it. There is
repression only where there is internal conflict. A repressed thought is not just one
that is unconscious: it is one that one must struggle to keep unconscious. One
must be conscious of a thought if one is to struggle to keep it unconscious. But, in
that case, that thought isn’t repressed, given that a thought must be unconscious
to be repressed. Thus the concept of repression is incoherent.
Here is an illustration. I discover that my best friend has nothing but mali-
cious intentions towards me. Because that knowledge causes me acute distress, I
repress it and struggle to keep it repressed. For me to try to keep that thought re-
pressed is for me to try to make the following proposition be true:

(*) my belief that my best friend has nothing but malicious intentions towards me
remains unconscious.

But if I am trying to make (*) be (or remain) true, then I am ipso facto aware of the
fact corresponding to the underlined proposition, in which case my awareness of
that fact isn’t unconscious and therefore isn’t repressed. Of course, what we said
about that particular awareness is true of any other candidate for repression. Thus,
nothing could possibly be repressed, and the concept of repression is thus an in-
coherent one.

To deal with Sartre’s argument we must remember two key points: consciousness is not
the same thing as awareness34, and objects of consciousness are not, at least not cate-
gorically, constituents of consciousness. But we must also keep a third point in mind:

34. More precisely, not all awareness is consciousness, even though, at least arguably, all
consciousness is awareness.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

there are two kinds of knowledge – knowledge by acquaintance and knowledge by

description.
As we saw a moment ago, for a thought to be unconscious is not for its owner to
be unconsciously aware of it: it is simply for its owner not to be aware of it. And re-
pressing a thought consists not in driving one’s awareness of it into unconsciousness,
but rather in one’s ceasing to be aware of it. One represses a thought not by burying it,
but by turning one’s consciousness away from it.
The next step in our argument must be taken slowly and carefully. There is nothing
puzzling about the fact that you can stop looking at some troubling spectacle, e.g. a
mangled body by the side of the road. But what is puzzling is that, if there is such a
thing as repression, one can deliberately not think about something that one wishes
not to think about. Here the analogy concerning the mangled body appears to break
down. (I say “appears to break down” because it doesn’t actually break down, as we’ll
soon see.) To deliberately not look at a hideously mangled body, one has to be con-
scious of what it is that one wishes not to look at. So deliberately not looking at the
body involves being conscious of its presence. But if one is conscious of some thought
that one wishes not to have in one’s consciousness, then in virtue of that very fact it is
in one’s consciousness. So how is repression possible?
A continuation of the story told a moment ago can help us answer this question.
The sight of the mangled body was so troubling to you that you wish never to have to
behold it again. Here is how you might go about carrying out this resolution. Because
you know that the body is in location X, you take care to avoid being in a place from
which X was visible. You thus know that there is something horrible at X – something
so horrible that, if you saw it, you would subsequently experience intolerable distress
– and you therefore take pains never henceforth to be in a position to see anything that
is in that place. In light of these points, consider the proposition:
(H1) at location X, there is a hideously mangled body and, were I to see that body, I
would subsequently feel intolerable distress.

In virtue of knowing H1, you know the truth of some proposition that describes the
grisly spectacle that you wish to avoid seeing. But when you see the body, you are di-
rectly conscious of that spectacle itself. In the one case (seeing the body), you have
knowledge by acquaintance; in the other (knowing the truth of H1), you have knowl-
edge by description. Knowing H1 involves having some kind of awareness of some
horror. But, for obvious reasons, that consciousness is psychologically more innocu-
ous than the kind of awareness one would have in virtue of seeing the body. This point

Incidentally, it doesn’t seem true that all consciousness is awareness, since it is by no means
clear whether an itch or a tickle is an awareness. In Kuczynski (2003) I argue that, indeed, not all
consciousnesses are awarenesses.
As it happens, Sartre himself (1943: 11) holds that not all consciousnesses have objects:
“Not all consciousness is knowledge [i.e. not every consciousness has an object] (there are states
of affective consciousness for example).”
 Conceptual Atomism and the Computational Theory of Mind

is easily generalized: there are many contexts where knowledge by description is psy-
chologically innocuous and where knowledge by acquaintance is not. In fact, knowl-
edge by description of a given fact is always more innocuous than knowledge by ac-
quaintance of that same fact (even though knowledge by description obviously isn’t
always innocuous tout court).
Of course, if you know the truth of H1, you haven’t yet repressed your knowledge
of the horrible spectacle at X. But a slight alteration of our story does describe an in-
stance of repression, and does show that the concept of repression is a coherent one.
When you first look at the hideously mangled body, you see that it is that of your be-
loved brother Tom. (Thus far, you haven’t repressed anything.) You thus know the
truth of the following proposition:
(H2) In place X lies the hideously mangled body of my beloved brother Tom.

You know that, were you to remain aware of the truth of H2, the ensuing distress
would be unbearable. Given your knowledge of this fact, and given also your knowl-
edge of H2, you know the truth of the following:
(H3) In place X, there is something or other that is so distressing that, were I to see it, I
would not be able to bear the subsequent feelings of distress.

You learn the truth of H3 by way of your knowing the truth of H2. But once you know
H3, you can drop your knowledge of H2. Here is an illustration of the principle at work
here. Suppose you know that Tim has red hair and, on that basis, know that somebody
or other has red hair. Even if you forget ever having met Tim, and thus cease to know
the truth of the singular proposition Tim has red hair, you do not on that account cease
to know the truth of the general proposition somebody or other has red hair. General
knowledge may outlive the singular knowledge from which it is derived. It is thus per-
fectly possible for you to continue to know H3 even if you cease to have knowledge of
H2. For similar reasons, you could cease to know that:
(H4) I know that, in place X, lies the hideously mangled body of my beloved brother
Tom

without ceasing to know that:

(H5) I know that, in place X, there is something so distressing that, were I to see it, I
would not be able to bear the subsequent feelings of distress.

Once again, a continuation of our story about red-haired Tim may come in handy. You
meet Tim at time t. For a few months after t, you know both that Tim has red hair and
also that at least one person has red hair. Moreover, you know that you know both of
these things. In other words, you know:
(T1) I know that Tim has red hair
 Chapter 18.  Fodor’s third argument for conceptual atomism 

and also
(T2) I know that at least one person has red hair.

Later you forget about Tim – you forget ever having known him. (You haven’t repressed
your knowledge of Tim: your memory of him has undergone garden-variety deterio-
ration.35) But you still know that at least one person has red hair. Under those circum-
stances, you couldn’t possibly know T1. (You couldn’t possibly know it because it
wouldn’t be true, given that you no longer had any idea who Tim was.) But you very
well could continue to know T2. For exactly similar reasons, you could cease to know
H4 without ceasing to know H5.
We can now begin to close the argument. Suppose, once again, that you know H4
– so suppose that you haven’t yet forgotten (or repressed) your knowledge of that truth.
In that case, you know (or, at least, are in a position to readily learn) that:
(H6) For some object O, I know that there is some singular proposition P that has the fol-
lowing characteristics. First, P is true. Second, P has the form the hideously mangled
body of O lies in place X. Third, my knowledge of P causes me intolerable distress.

Remember what we said about H5 – one can know that proposition and not know H4,
even if one learned H5 by way of learning H4. Of course, if one were to forget H4, H6
wouldn’t be true any more. (For if one were to forget H4, then one would no longer
know P’s identity, and the third condition described by H6 therefore wouldn’t be ful-
filled.) But what we said about H5 would be true of the following variant of H6:
(H7) For some object O, I know that there is some singular proposition P that has the
following characteristics. First, P is true. Second, P has the form the hideously
mangled body of O lies in X. Third, were I to know P’s identity, I would experience
intolerable distress.

Some clarificatory remarks (again involving red-haired Tim) might be in order. If you
know that Tim has red hair, you know (or are in a position to readily learn) that, for
some x, some singular proposition of the form x has red-hair is true. And you can
continue to know that, for some x, some singular proposition of that form is correct,
even if you have long since forgotten about Tim (and any other specific red-haired
individual). Similarly, even if you learned H7 by way of learning H4, you could cease
to know the latter without necessarily knowing the former.

35. For the sake of expository simplicity, I am supposing that there is such a thing as garden-
variety forgetting. Freud (1998: Chapter 7) denies this, saying that, setting aside cases of brain-
damage and brain-deterioration, what we think of as cases of mere forgetting consist in our
minds deliberately suppressing or recoding existing memory. What we think of memory-dele-
tion, says Freud, is really memory-relocation of memory-corruption. Nietzsche makes a similar
point in The Genealogy of Morals. I am inclined to agree with Freud and Nietzsche about this,
and I think that their view is borne out (albeit weakly) by the fact that an analogue of it is true of
computers.
 Conceptual Atomism and the Computational Theory of Mind

If you know H4, then you know (or are in a position to learn forthwith) that:
(H8) I know that I know that, in place X, lies the hideously mangled body of my be-
loved brother Tom.

And if you know H7, then you know (or are in a position to learn forthwith) that:
(H9) I know that, for some object O, I know that there is some singular proposition P
that has the following characteristics. First, P is true. Second, P has the form the
hideously mangled body of O lies in X. Third, were I to know P’s identity, that would
cause me intolerable distress.

Given obvious extensions of the points already made, H9 could be true even if H8 was
not. If you know H9, you know that there is some proposition that you don’t want to
know. In other words, you know a proposition that describes a proposition that you
don’t want to know. You are acquainted with a proposition that describes a proposition
that you are not acquainted with and whose identity you wish not to know.
There are many cases where somebody knows what it is that he doesn’t want to
know. For example, most people know that they don’t want to know about their par-
ents’ sexual habits. Thus, most people know that, for some m and some n, they don’t
want to know a true proposition of the form my parents frequently have sex in manner
m or my parents have sex on n occasions per year or my parents have sex in manner m
on n occasions per year. Thus, people know descriptions of propositions whose exact
identities they do not wish to know. They know by description just what it is that they
don’t want to know by acquaintance.
Supposing that you know H9, you know that there is some truth that you very
much don’t want to know. You know some proposition P* that describes some proposi-
tion P whose exact identity you don’t want to know. Under these circumstances, you
could, and probably would, struggle to not find out what P is. You would have enough
descriptive knowledge of P to know when you were on the brink of discovering P’s
identity. Therefore, given your knowledge that a knowledge of P’s identity would cause
you intolerable distress, you would feel intense anxiety when you were likely to acquire
that knowledge.
The principle at work here can be understood in terms of a point that we made a
moment ago. Somebody who doesn’t know any truth of the form my parents often have
sex in manner m can obviously know when he is in a situation where he is bound to
acquire such knowledge, and (supposing that he wishes not to acquire such knowl-
edge) will feel anxiety if he is in such a situation. (If you are leafing through your
mother’s diary, you know that you are likely to learn a truth of this exact form, and this
knowledge is bound to make you feel anxious.) Similarly, somebody who doesn’t know
any specific proposition of the form were I to know proposition P, I would feel intolerable
distress may know enough to know when he is about to learn the truth of such a prop-
osition. For such a person may be acquainted with some proposition that describes P.
 Chapter 18.  Fodor’s third argument for conceptual atomism 

Indeed, such a person may in fact be acquainted with some proposition singles out P
(i.e. that is satisfied by P and P alone).
The phenomenon of resistance is easily understood in these terms. It is an empiri-
cal fact that people often fiercely resist drawing conclusions that, given the data at their
disposal and given their level of intellectual acumen, should be patently obvious to
them. So long as we don’t forget the distinction between knowledge by acquaintance
and knowledge by description, there turns out to be nothing paradoxical about this
pervasive and undeniable phenomenon. You could obviously resist being put in a situ-
ation where you would know that you were on the brink of learning some fact about
your parents’ sexual behavior. This would be the case if you knew some proposition P
describing some proposition P* such that (i) P* concerned your parents’ sexual con-
duct, (ii) you knew that a knowledge of P* would disturb you, and (iii) you knew that,
given the circumstances, you were likely to learn P*. Thus, the concept of resistance
doesn’t have to be understood in terms of a conflict between the conscious and the
unconscious (even though, I am trying to show, it can sometimes be so understood).
That concept can be understood in terms of facts whose existence both Freudians and
non-Freudians will concede.
In order to close the argument, we must recall a point that we made earlier. There
are two ways that a propositional attitude can be unconscious. It can be unconscious
in the sense that, even though you are not conscious of it at the moment, you could
readily become so. Consider, for example, your belief that 1+1=2. A few seconds ago
you were not (let us suppose) conscious of that belief. But this isn’t because you were
emotionally resistant to becoming aware of it; it is simply because your mind was oc-
cupied with other things. Freud describes propositional attitudes that are unconscious
in this sense as “preconscious” and also as “topographically unconscious.”
But, if Freud is right, there is another way that a propositional attitude can be un-
conscious: a propositional attitude can be unconscious because you don’t want to be-
come conscious of it. Such propositional attitudes are “dynamically unconscious” or
“repressed.”
Given what we’ve said in this section, we can give a precise definition of the term
“repressed” – a definition that, if correct, invalidates Sartre’s criticism of the concept of
repression. An attitude towards proposition P is repressed if (a) its owner is not con-
scious of having that attitude, (b) its owner is conscious of some proposition P* that
describes P, and (c) he knows that he doesn’t want to know, i.e. be acquainted with, the
proposition that P* describes. If these three conditions are met, then the person who has
that attitude will obviously resist becoming acquainted with P. Given his knowledge of
P*, the subject knows when it is that he is likely to learn the identity of the proposition
described by P*, and he also knows that, were he to do so, he would suffer intolerable
distress. Hence, the subject will struggle not to acquire that knowledge; he will “resist”
efforts to impart it to him, just as you would resist efforts on the part of others to tell you
about your parents’ intimate habits. (One might think that, in order for condition (c) to
be met, the subject must know the identity of the proposition described by P*. This is not
 Conceptual Atomism and the Computational Theory of Mind

the case. For (c) to be met, the subject need only want it to be the case that he doesn’t
become aware of any proposition uniquely described by P*. Thus, a failure to understand
the de dicto/de re distinction underlies the view that one must be acquainted with the
proposition described by P* to want to not know it. In fact, as we are seeing, that failure
appears to underlie much criticism of the concept of repression.)
Of course, the two situations are not perfectly analogous. In the one case, you are
trying not to become conscious of knowledge that you already have. In the other case,
you are trying not to acquire knowledge that you don’t have at all. But there is nothing
paradoxical about the concept of a person’s trying not to be conscious of a piece of
knowledge that he already has. A brief review of some earlier points may be helpful
here. Since propositional attitudes are not constituents of consciousness, they can be
“conscious” only in the sense of being objects of consciousness. Nothing that is an object
of consciousness must be an object of consciousness. So propositional attitudes can be
unconscious. (Indeed, they are almost always unconscious: it is only on rare occasions
that you are conscious of your belief that 1+1=2.) Given that propositional attitudes can
be unconscious, and given also that you might be conscious of a proposition that de-
scribes an unconscious propositional attitude of yours in such a way as to make you not
want to become conscious of that attitude, it follows that there are “dynamically uncon-
scious”, or “repressed”, propositional attitudes, and it also follows that you will be emo-
tionally resistant to learning some facts about your own self. So there does not appear
to be any a priori reason why Freudian concepts such as repression and resistance
couldn’t be instantiated. Further, given how obvious it is that people refuse to draw
conclusions that they ought to draw, given the extent of their knowledge and intelli-
gence, it is hard to see how Freud’s basic views could be entirely wrong.
 chapter 19

Some arguments for the Symbolic Conception

of Thought

We’ve seen some reasons to hold that SCT is false, and in Chapters 22–23 we will find
confirmation for this view. But Fodor (1975) provided five razor-sharp arguments on
behalf of SCT, and we must take these into account before we come to any definitive
conclusions as to whether SCT is true or false.

Argument #1 for SCT: the resemblance between thought and language

Thought resembles language. Sentences can be true or false; the same is true of
thoughts. Sequences of sentences can be valid or invalid; the same is true of se-
quences of thoughts.“Socrates” picks out Socrates; the same is true of my concept of
Socrates. Given how much our thoughts resemble linguistic symbols, it seems rea-
sonable to say that our thoughts are such symbols.

Here is what this amounts to: anything that is true, false, or otherwise representational,
is ipso facto symbolic. But if we take that position, then the thesis that we think in
symbols becomes the platitude that our thoughts are true, false, and otherwise
representational. But the thesis that thoughts are symbol-tokens is vacuous unless it is
taken to mean that thoughts and symbol-tokens have properties in common other
than the properties of being true, false, and otherwise representational. Given what
we’ve seen, there is no evidence that thought and language have any significant com-
monalities apart from those just mentioned. We will find confirmation for this posi-
tion when, in Chapters 22 and 23, we discuss Evans’ distinction between “conceptual”
and “non-conceptual” content.

Argument #2 for SCT: the systematicity of language

Languages are systematic: a natural language can express any viable permutation
of the concepts expressed by any sentence belonging to it. If a language can ex-
press the proposition Tom loves Mary, then it can also express the proposition
Mary loves Tom. To take another example: if a language can express the proposi-
tion: supposing that they were anatomically capable of playing the piano, penguins
 Conceptual Atomism and the Computational Theory of Mind

would find Bach’s contrapuntal compositions easier to perform than Schumann’s

non-contrapuntal compositions, then it can also express the proposition: supposing
that he were anatomically capable of playing the piano, Schumann would find Bach’s
non-contrapuntal compositions easier to perform than penguins’ contrapuntal com-
positions. Languages are “systematic.”
Like language, thought is systematic. This property of thought is easily ex-
plained if it is supposed that we think in a language. The systematicity of thought
provides some evidence that we think in a language.

To speak English is to have internalized various semantic rules. A precondition for-

one’s being able to manipulate these rules is that one’s thought be systematic. This is a
matter of logical, as opposed to causal, necessity. Your understanding “Tom loves
Mary” consists in your being able to compute the identity of the proposition meant by
that sentence (or, strictly speaking, occurrences thereof) on the basis of the relevant
semantic rules. The very act of applying those rules to a new sequence of expressions
is itself an instance of the systematicity of thought.
Consider the expression “loves.” That expression must be defined contextually: an
utterance of Fx loves yG is true iff x loves y. (Of course, this point is not self-evident. In a
moment, we will see why it is true.) Since “x” and “y” are variables, understanding the
rule just described consists in one’s being able to substitute different constants for those
variables. It involves being able to understand the result of substituting “Bob” for “x”
and “Samantha” for y; and, by the same token, it involves being able to understand the
result of substituting “Samantha” for “x” and “Bob” for y.
In response to this, one might say that it is possible to understand “loves” without
being able to understand different sentences of the form Fx loves yG and, consequently,
that an understanding of that expression is not to be understood in combinatorial
terms. But such a position would be most counterintuitive. For the sake of argument,
suppose that Smith can understand only one substitution-instance of the formula Fx
loves yG – he can understand “Bob loves Mary”, but not “Mary loves Bob”, or “Harry
loves Ethel.” Let us also suppose, the words “Bob”, “Mary, “Ethel”, and “Fred” are in
Smith’s lexicon. In that case, Bob knows that the sound “BOBLOVESMARY” means
Bob loves Mary. But he isn’t able to parse that sentence; and what he understands is
therefore “BOBLOVESMARY”, not “Bob loves Mary.” From his viewpoint, “loves” is
no more a “part” of that sentence that “BLOVE” or “SMA” or “OBLO.” Smith cannot
distinguish semantic from phonetic constituency. Given this, it is hard to make the
case that “loves” is in his lexicon in the sense in which it is in your lexicon.
We said a moment ago that “loves” must be defined contextually. What is our argu-
ment for this? We just provided an implicit argument for it. Let us now make that argu-
ment explicit. A non-contextual definition – e.g. “loves” denotes the relation of loving
– wouldn’t give the syntactic properties of “loves”; it wouldn’t explain why “Tom loves
Mary” is meaningful, whereas “Tom, the relation of loving, Mary” is not. To understand
an expression is to know, not only what it refers to (supposing that it refers to anything),
but also how it syntactically combines with other expressions. Contextual definitions
 Chapter 19.  Some arguments for the Symbolic Conception of Thought 

provide this information; non-contextual definitions do not. (So far as non-contextual

definitions do provide such information, it is because they are being tacitly equated with
contextual definitions. If I say “the property of loving” denotes the property of loving,
you tacitly understand that “the property of loving” is not interchangeable with “loves.”
You know, for example, that “the property of loving is instantiated” is meaningful,
whereas “loves is instantiated” is not. Obviously the combinatorial properties of “the
property of loving” are not given by a denotative definition of that term, and must there-
fore be given by a contextual definition. So your knowledge of the meaning of “the prop-
erty of loving” is identical with knowledge of a contextual definition.) To understand an
expression is to know its combinatorial, and not merely its referential, properties. Hav-
ing such knowledge involves knowing how to combine it with other expressions, and a
prerequisite to one’s having such knowledge is that one’s thinking be systematic. So the
systematicity of language is to be understood in terms of that of thought, and not vice
versa. Given that language is systematic, it is viciously regressive to suppose that thought
is linguistic, and the systematicity of thought thus bars us from accepting (SCT). We
already saw why the standard SCT-counter-response is untenable.
We just saw that a certain amount of cognitive systematicity is needed to grasp any
contextual definition, e.g. FP or QG is true if P is true or Q is true. By the same token,
given a certain amount of cognitive systematicity, a group of people can create seman-
tic rules that build systematicity into their modes of communicating with one another.
This confirms our suggestion that linguistic systematicity is a special case of cognitive
systematicity, and not vice versa.
Of course, no English-speaker can understand every meaningful permutation of any
group of English expressions. But, as Chomsky (1988) plausibly argues, that has to do
with issues relating to performance, not competence. Also, once the linguistic rules de-
scribed in the last paragraph have been activated, there is no limit at all as to how com-
plex sentences of that language can be. Given contextual definitions like the one just de-
scribed, there are sentences of English that are too long for any person to understand but
that are demonstrably true. So whether or not people are able to understand an English
sentence that is 87,000 words-long, it remains a fact that the systematicity of English is a
reflection of the corresponding cognitive simplicity of its users (and creators).1
In conclusion, the systematicity of language is to be understood in terms of the
systematicity of thought, and it is therefore viciously circular to attempt to understand
the systematicity of thought in terms of that of language.

1. Fodor and Pylyshin (1988) argue that CTM is preferable to a doctrine known as “connec-
tionism” on the grounds that connectionism is harder to reconcile than CTM with the systema-
tic character of thought. Whether connectionism is a correct or otherwise meritorious doctrine
falls outside the scope of the present work. But given what we’ve said, it is clear that CTM doesn’t
explain the systematic character of thought at all and that it therefore doesn’t explain it any bet-
ter than connectionism.
 Conceptual Atomism and the Computational Theory of Mind

A second argument against argument #2

The statement that languages are necessarily systematic is false. We can imagine a lan-
guage L that has a single, semantically simple symbol that means Mary hates Bob, but
that doesn’t have distinct symbols for Mary, Bob, or the relation of hitting. In that case,
even though L can express the proposition Mary hates Bob, it cannot necessarily ex-
press the proposition Bob hates Mary.
There doesn’t seem to be any reason in principle why a language might not consist
only of semantically simple expressions that expressed whole propositions. Such a lan-
guage would not (at least not necessarily) be systematic at all. So given only the sup-
position that a representational medium is a language, it doesn’t follow that it is sys-
tematic. Thus, given only the supposition that we think in a language, we don’t have
any explanation as to why thought is systematic.
Of course, given the supposition that we think in a language along with the sup-
position that the language in question is systematic, we do have such an explanation of
the fact that thought is systematic. But it must then be explained on empirical grounds
why that internal language is systematic. And in that case, the supposition that we
think in a language explains the systematic character of thought only to the extent that
there is some other, independent explanation of that fact. Therefore, the systematicity
of thought is to be explained in terms of that other, independent explanation, and the
supposition that we think in a language becomes completely gratuitous.
As we’ve just seen, given only that we use languages, it doesn’t follow that we use
systematic languages. So it is an open, empirical question why the languages that we
use are systematic. Here is one possible answer to that question. We cannot say what
we have to say through a language that is unsystematic; an unsystematic language
wouldn’t do justice to our thoughts, just as a palate consisting only of two colors
wouldn’t do justice to Cezanne’s artistic vision. For this reason, we prefer language L to
language L* if ceteris paribus L is more systematic than L*. Given this preference of
ours, we tend to use systematic modes of expressions.
Unless thought were systematic, it is hard to see why ceteris paribus systematic
languages would do a better job than non-systematic languages of expressing our
thoughts. Once again, it turns out that the systematicity of language is to be under-
stood in terms of that of thought.

Argument #3 for SCT: the productivity of language

If a public language (e.g. English or Spanish) can express the proposition snow is
white and also the proposition grass is green, then it can express propositions con-
sisting of those propositions, e.g. grass is green and snow is white, grass is green or
snow is white, and so on. Languages are productive: languages can express molecu-
larizations of the sentences belonging to them. Thought has a similar property. If
 Chapter 19.  Some arguments for the Symbolic Conception of Thought 

a person can think grass is green and snow is white, then he can think grass is green
and snow is white. Given that language is productive, the productivity of thought
is easily explained by supposing that we think in a language.

First of all, the productivity of language seems to be an aspect of its systematicity.2

Productivity is to whole sentences what systematicity is to sub-sentential
expressions. Given that productivity is a kind of systematicity, it follows that every-
thing we just said about the latter is true of the former and, therefore, that Fodor’s third
argument has the same defects as his second.
Also, the premise of argument #3 is false or, in any case, is only contingently true.
As we saw, there is nothing absurd in the supposition that a language L might contain
a primitive expression meaning Bob loves Mary, but no expression meaning Mary loves
Bob. Given obvious extensions of what we said a moment ago, in connection with ar-
gument #2, it follows that argument #3 does not go through.

Argument #4 for SCT: grasped but unaffirmed propositions

When you say “either snow is white or penguins are mammals” or “if penguins are
mammals, then some mammals live in Antarctica”, you are not asserting that penguins
are mammals. Not every assertible constituent of an utterance is itself asserted.
Similarly, not every believable constituent of a thought is itself believed. You
can believe either snow is white or penguins are mammals without believing either
disjunct. Given that language has a corresponding property, this property of
thought is plausibly explained by assuming that we think in a language.

Given what we’ve said so far, the obvious criticism of this argument is that this property
of language is to be explained in terms of the corresponding property of thought, and not
vice versa. I believe that this criticism is cogent, and that there are special reasons to think
that this particular argument of Fodor’s is guilty of the error just cited.
Before we develop our criticism of argument #4, we need to clarify some points
concerning the concept of sentential-constituency. It would be deeply arbitrary to say
that the occurrence of “snow is white” in
(*) “snow is white and grass is green”,
were in a different grammatical category from its counterpart in:
(**) “snow is white or grass is green.”
In particular, it would be arbitrary to say that the occurrence in (*) did have assertoric
force, whereas its counterpart in (**) did not. (*) and (**) make different statements, of
course. But that isn’t because “snow is white” has assertoric force in one of them but

2. Fodor and Pylyshin (1988) make this point.

 Conceptual Atomism and the Computational Theory of Mind

not the other. It is because of the specific meanings of “or” and “and.” For exactly simi-
lar reasons, the occurrence of “snow is white” is no more asserted in “it is true that
snow is white” than it is in “it is false that snow is white.”
What this shows is that any sentence that is a constituent of another sentence ipso
facto lacks assertoric force – even if, because of the verbiage in which that occurrence
is embedded, the proposition meant by that occurrence is affirmed. A corollary is that
any sentence-occurrence that has assertoric force automatically forms a sentence (or
sentence-token, to be more precise) unto itself. More generally, any sentence-occur-
rence that has any kind of force – be it assertoric, imperatival, interrogative, exhorta-
tory…– is a sentence unto itself, and is not a constituent of another sentence.
There is one last preparatory point to make. Only sentences can be asserted. You can’t
assert “Socrates” or “red”, unless your utterance is elliptical for a complete sentence, e.g. “the
color of my new shirt is red” or “Socrates is my favorite philosopher.”
In light of these points, let us ask: what does it mean to say that not every assertible
constituent of a sentence is itself asserted? Given that any constituent of a sentence is
ipso facto unasserted, and given that nothing other than a sentence can be asserted (i.e.
can have assertoric force), it means nothing more, and nothing less, than that there are
complex sentences. It means, in other words, that there are sentences that are com-
posed of other sentences.
Similarly, to say that thought has a property parallel to the one just described is to
say nothing more, and nothing less, than that there are complex thoughts – thoughts
that consist of other thoughts.
This brings us to what appears to be (and, we will see, actually is) a fundamental
difference between thought and language. As we’ve discussed, there is a definite sense
in which a sentence has minimal units of significance. But, by all appearances, the
same is not true of thought. (In this context, “thought” refers not just to discursive
mentation, but to anything mental that is representational.) Your current visual per-
ception of this book doesn’t decompose into minimal units of significance. In any case,
if it does so decompose, that would have to be demonstrated by an extraordinarily
non-obvious line of argumentation.
Fodor aspires to explain the previously discussed property of thought (that not
every constituent of what is believed is itself believed) in terms of the corresponding
property of language (that not every assertible constituent of an assertion is itself as-
serted). This is certainly a reasonable hypothesis. But it is not an independently cor-
roborated one, i.e. it isn’t borne out by data other than the data that motivated it. After
all, when we look at mental activity other than verbal activity, we are struck by the dif-
ferences, not the similarities, between mental and linguistic representation.
An advocate of SCT could respond by saying that sense-perceptions actually do have
a sentence-like structure, even though their phenomenologies may suggest otherwise.
But that position is ad hoc. Also, it is question-begging in this context, since Argument
#4 is meant to show that SCT is right, and therefore cannot assume its truth or that of any
 Chapter 19.  Some arguments for the Symbolic Conception of Thought 

doctrine whose truth is predicated on that of SCT. (In Chapters 22 and 23, we will argue
that the position just described not only seems false, but actually is so.)
But it isn’t just sense-perceptions that are prima facie counterexamples to Argu-
ment #4. Thoughts themselves appear to be such counterexamples. One can say “Soc-
rates” (without its being elliptical for a whole sentence) or “or” or “because.” Of course,
such one-word utterances are ill-formed, but there is nothing impossible about them.
(Indeed, they are a pre-condition for multi-word well-formed utterances.) But one
cannot think just Socrates. One must think Socrates was bald or Plato admired Socra-
tes.3 This suggests that thoughts are not built up out of discrete parts corresponding to
the ultimate constituents of sentences. A thought that Socrates admired Plato there-
fore seems not to have a decomposition comparable to that of the sentence “Socrates
admired Plato.” This is a point we will soon develop.
An advocate of SCT might respond by saying: “If SCT is right, then mental repre-
sentations (i.e. thoughts and perceptions) do have a structure similar to that of a sen-
tence; and if SCT is right, then you can just think Socrates.” But, for reasons analogous
to those discussed a moment ago, such a response would be question-begging.
There is another problem with Argument #4. What does it mean to say that:
(a) “snow is white”
is a “constituent” of
(b) “either snow is white or grass is green”?
The obvious answer is to say that any token of the latter has a token of the former as a
component. But what does it mean to say that a token t of (a) is a “component” of some
token t* of (b)? It obviously won’t do to say that t is a spatiotemporal part of t*. The un-
derlined part of:
(c) either snow is white or grass is green”
is a spatiotemporal part of that sentence-token. But it isn’t a part of it in any semanti-
cally significant sense; it isn’t a part of (c) in the same sense as the bold-faced portion.
When we say that the bold-faced part is a constituent of (c), we mean that the
derivation-tree for the former is a part of the derivation-tree for the latter. We mean,
in other words, that the structured set of semantic rules that assign meaning to the first
is a subset of the set of semantic rules that assign meaning to the second. We’ve already
seen how, where public languages are concerned, the concept of a semantic rule is to
be understood in psychological terms. The same is true of the concept semantic con-
stituency (i.e. of one expression’s being a constituent of another), given that this con-
cept is to be understood in terms of that of a semantic rule. This suggests that semantic
constituency is to be understood in terms of facts about thought, and not vice versa. In
its turn, this confirms our larger point that, because the central concepts of semantics

3. Robert Brandom (1998) makes this point.

 Conceptual Atomism and the Computational Theory of Mind

are to be understood in psychological terms, it is impossible to establish enough mean-

ingful parallels between thought and language to warrant an acceptance of SCT with-
out trivializing that doctrine.

Argument #5 for SCT: induction and the simplicity-metric

Let us consider yet another argument given by Fodor on behalf of SCT. This argument,
it will be seen, is in a different vein from the four just considered. We will refer to it as
“Fodor’s argument from induction” (FAI):
(FAI) Obviously much of our thinking consists in the making of inductive infer-
ences. But Goodman showed that any data that warrants an intuitively rea-
sonable inductive inference also warrants an intuitively unreasonable one. Let
D be the fact that every emerald observed before time t has been green. Given
D, it is reasonable to infer that:
(G) all emeralds are green.
But D confers no less weight on that reasonable hypothesis than it does on the
unreasonable hypothesis that:
(B) all emeralds are grue,
where “grue” is defined thus: x is grue iff x is green before time t and blue thereafter.
Nonetheless, there is presumably there is some legitimate, as opposed to pure-
ly subjective, reason to prefer (G) to (B). In general, notwithstanding Goodman’s
correct point, there must be some reason to reject those hypotheses that we char-
acterize as “counter-inductive” in favor of those that we characterize as “induc-
tively reasonable.” (I am taking it for granted that somebody who prefers (G) to
(B), given (D), is ceteris paribus better at inductive reasoning than somebody who
has the opposite preference.) The difference lies in the fact that, given our various
background beliefs, accepting (G) would lead to a simpler and more integrated
body of beliefs than would accepting (B). Given those beliefs, considerations of
simplicity warrant our believing (G) as opposed to (B), even though by itself (D)
confers no more weight on (G) than it does on (B).
Given data D1…Dn, hypothesis H is ceteris paribus preferable to hypothesis
H* if acceptance of H is theoretically simpler than acceptance of H*. (Of course,
theoretical simplicity must be distinguished from psychological simplicity. Theo-
retical simplicity is a property of the relationship between hypothesis and data.
Psychological simplicity is a property of the relationship between belief in a hy-
pothesis and belief in data.) We seldom make bent inductions: when our induc-
tions are wrong, it is (almost) never because they are in the same category as (B).
This means that ceteris paribus we prefer simple to complex hypotheses. This in
turn means that we have some way of comparing hypotheses in respect of how
 Chapter 19.  Some arguments for the Symbolic Conception of Thought 

simple they are (or, more precisely, in respect of how simply they can be assimi-
lated into our existing beliefs). Thus, encoded in our thinking is some kind of
“simplicity metric”, without which we would make bent, wrong inductions.
Simplicity, in the relevant sense, is to be understood in terms of the syntaxes of
sentences. Suppose that L is a scientifically viable method of coding hypotheses into
sentences and that, given any proposition P, there is a unique sentence S(P) of L that
encodes P. [To facilitate our exposition of Fodor’s argument, let us prescind from the
fact that, in this context, use of the term “scientifically viable” might constitute vi-
cious circularity, given that it is a near synonym of “inductively viable.”] In that case,
H’s being simpler than H* is expressed in (S)H’s having some syntactic property that
S(H*) does not. In general, there is some syntactic property PHI such that, given any
two hypotheses x and y, x is simpler than, and thus ceteris paribus preferable to, y just
in case S(x) has PHI to a greater degree than S(y).4

Everything said up until the very last paragraph of (FAI) is extremely reasonable. For
reasons that I give elsewhere5, I believe that much (though obviously not all) of that
material is wrong. But let us suppose, if only for argument’s sake, that everything said
prior to the last paragraph of (FAI) is correct. Does it follow that simplicity is to be
understood in terms of syntax?
No. Let us start with a point that might initially seem off-topic. Obviously people
can articulate at least some of their beliefs (e.g. I can express my belief that it is raining
by saying “it is raining”). There is thus a “semantic characterization” of many true
propositions. (More exactly, given a language L that can express a given proposition,
there is a semantic characterization of that proposition in L.)
But it doesn’t follow that the concept of assenting to the proposition that it is rain-
ing is to be understood in semantic or otherwise linguistic terms. Maybe it is so to be
understood (though we have seen reason to believe otherwise). Indeed, that is what
Fodor would say. But given only that we can express our thoughts in language, i.e.
given only that there is “semantic characterization of thought”, it doesn’t follow that
our thoughts are themselves linguistic expressions.

4. Fodor (1975: 39-40) writes: “[S]o far as anyone can tell, simplicity metrics must be sensitive
to the form of the hypotheses that they apply to, i.e. to their syntax and vocabulary. That is, so far
as anyone can tell, we get an a priori ordering of hypotheses [a simplicity metric] only if we take
account of the way in which the hypotheses are expressed. We need such an ordering if we are
to provide a coherent account of the order in which values of P are selected in the concepto-lear-
ning [and induction-forming] situation. But this means that a theory of concepto-learning [and
induction-formation] will have to be sensitive to the way that the organism represents its hypo-
theses. But the notion of the organism representing its hypotheses in one way or another (e.g. in
one or another vocabulary or syntax) just is the notion of the organism possessing a representa-
tional system.”
5. In Kuczynski (2007b).
 Conceptual Atomism and the Computational Theory of Mind

Similarly, supposing that there is a syntactic characterization of inductive reasona-

bleness, it doesn’t follow that our being inductively competent consists in our being
sensitized to sentences of some kind. What follows is only that this competence can be
given sentential expression.6
We must remember our earlier points about syntax. To say that S has a certain
syntax is to make a statement about the semantic rules that assign it meaning. So to say
that there is a syntactic characterization of inductive reasonableness is to say that there
is a possible language L satisfying the following two conditions:
(i) L is rich enough to express any inductively sound argument.
(ii) Supposing that S1…Sn are the sentences, or sequences of sentences, belonging
to L that express inductively sound arguments, there is some property PHI
such that the derivation tree of Si has PHI just in case 1≤i≤n.
Thus, sensitization to syntax is sensitization to facts about how sentences are assigned
meaning.
So far as one’s thinking is driven by facts about how expressions of language L are
assigned meaning, one’s thinking is about L, and not in L. So supposing that we think
in a language, it would be viciously regressive to say that our thinking were driven by
a sensitivity to facts about how its expressions were assigned meaning. Since sensitiza-
tion to syntax is sensitization to facts about how sentences are assigned meaning, it
follows that, so far as we think in a language, our thinking is not driven by the syn-
taxes of its expressions.
The only way to block this would be to say that, where that internal language is
concerned, the term “syntax” has a meaning entirely different from the homonym of
that term that is used in connection with public languages. But in that case, the state-
ment “thinking, and therefore the making of inductions, is a syntax-driven affair”
ceases to have any identifiable meaning.

6. In the argument being considered, Fodor seems to be alluding to Carnap’s (1950) efforts to
formalize induction. It is worth noting that Carnap’s system, along with all other efforts to for-
malize induction, have radically counter-intuitive results. See Pap (1962: Chapters 11-13) for a
discussion of this. For this reason, it is no longer widely held that induction can be formalized.
It should be pointed out that, given only that induction cannot be formalized, it doesn’t
follow that there is no analytic, non-trivial characterization of inductive rationality. Let L be any
language that is capable of expressing all inductively reasonable inferences. To say that induction
cannot be formalized is to deny that, for any sentence S of L that expresses an inductively reaso-
nable inference, it is a theorem of L’s semantic rules that S has that property.
But we know that the category of formal truth is a small sub-category of the category of
analytic (i.e. strictly conceptual, non-empirical) truth. This is to be expected, given that the
concept of formal truth must be understood in terms of concepts (such as theorem and conse-
quence) that are themselves to be understood in terms of the concept of analytic truth. So given
only that induction cannot be formalized, it doesn’t follow that induction defies rational analysis
any more than mathematics.
 Chapter 19.  Some arguments for the Symbolic Conception of Thought 

For reasons discussed earlier, it would be no use to counter-respond by saying that

our thinking is syntax-driven in that it is morphology-driven, and morphology can be
aligned with syntax.
 chapter 20

A positive argument against SCT

We have seen some reasons to doubt the cogency of Fodor’s arguments on behalf of
SCT. But given only that Fodor’s arguments don’t go through, it doesn’t follow that
SCT is wrong. But there are some reasons to think that SCT is in fact wrong.
The standard argument against SCT is this. Given any expression belonging to a
public language, what that expression means is derivative of human thought. Therefore
it is viciously regressive to suppose that thought is itself linguistic.1
I believe that the argument just given is cogent. (In a moment we will consider
some possible objections to it.) But the premise of that argument is liable to be misun-
derstood. Given only that word-meaning is parasitic on human thought, it doesn’t
follow that Grice’s analysis of meaning is correct; in other words, it doesn’t follow that,
if E has meaning M, that is because people mean M by E. Nor does it follow that hu-
man thought directly fixes what expressions mean. (There are very long sentences of
English that have definite meanings but that have never been contemplated. So far as
human thought fixes the meanings of such sentences, it does so by way of fixing the
meanings of their constituents.) Nor, finally, does it follow that human thought single-
handedly fixes what expressions mean. (Putnam is right: on Twin-Earth, tokens of
“water” refer to XYZ, not H2O. But, as Putnam (1975) argued2, Bob and Twin-Bob
have exactly the same thoughts. ) What follows is that what people think is part of what
fixes what expressions mean. (How it fixes it, and to what extent, is another question.)
This is obviously a reasonable point and, if it is right, it is viciously regressive to sup-
pose that we think in a language.
The SCT counter-response is to say that the meanings of Mentalese symbols are
not fixed by thought. But in that case, an opponent of SCT would say, it becomes ques-
tionable whether the so-called symbols of Mentalese deserve to be referred to as such,
given how different they are from things that we know to be symbols.
At this point, communication between advocates and opponents of SCT tends to
stop. But my feeling is that, were it to continue, the debate might well take the follow-
ing direction. Advocates of SCT would respond to the criticism stated a moment ago

1. See Dennett (1978: Chapter 6), Lycan (1984: 237), Kuczynski (2002, 2005), Searle (1984, 1992).
2. Putnam (1996) later changed his position on this. As we’ve seen, Burge (1979, 1982) and
others have argued that extensions of Putnam’s arguments for semantic externalism could esta-
blish the truth of content-externalism. Putnam came to regard these arguments as cogent, and
thus came to be a content-externalist: he came to hold that Bob and Twin-Bob are having diffe-
rent thoughts. As we’ve seen, those arguments are not cogent and, given a few basic distinctions,
content-externalism loses any plausibility that it might otherwise have.
 Conceptual Atomism and the Computational Theory of Mind

by saying that they are indeed extending the concept of a symbol, but that scientific
progress consists in making principled extensions of this sort: surely physics wouldn’t
have made much progress if physicists insisted on operating within the confines of the
layperson’s concepts of temperature, mass, energy, and so on.
The anti-SCT counter-response would be to say that SCT has over-extended the
concept of a symbol and that what it is doing in connection with the concept of a sym-
bol isn’t comparable to what physics has done in connection with concepts of tem-
perature and mass.
But at this point the debate is quite murky. Who is to say when an extension of a
concept (or expression) is justified and when it isn’t? I would like to make an attempt
to break this stalemate.

The different meanings of the word “language”

There are three important distinctions that relate to the study of language. (When I say
“important”, I don’t necessarily mean “valid.” As we will see, one of them is quite spurious.
The other two, however, are legitimate.) They all bear a certain resemblance to one another,
and are seldom carefully distinguished. But in this context they must be distinguished.
Saussure (1966) distinguished “langue” and “parole.” The former refers to language
per se, and the latter refers to the use of a language.
David Lewis (1975) distinguished between LANGUAGES and languages*. (He used
a different notation to mark this distinction.) LANGUAGES are sets of purely platonic
objects: function-theoretic pairings of properties, individuals, and the like, with physical
types or tokens. It is evident that such pairings are themselves platonic entities, and thus
have no spatiotemporal co-ordinates. Languages*, on the other hand, are constituted by
patterns of behavior and thought – by people producing certain noises and ink-marks in
response to various stimuli. According to Lewis, the word “language” is ambiguous be-
tween LANGUAGES and languages*; and the word “meaning” is comparably ambigu-
ous, sometimes referring to a purely function-theoretic relation (e.g. a function that as-
signs truth to an utterance of “that cat is mangy” exactly if there is a unique contextually
salient cat x, and x is mangy) and other times referring to a psychological phenomenon
(as in “what I meant is that you should buy some new clothes”).
By way of anticipation, Saussure’s distinction is legitimate, whereas Lewis’ is not.
Let us now discuss the third distinction (one that we will find to be valid and impor-
tant) – that between distinct, but historically connected, languages.
If we think of languages as sets of semantic rules, then the referent of the word
“English” in 1980 is not identical with the referent of the homonymous word in 2006,
thanks to (for example) the introduction of various new terms relating to the internet.
At the same time, there is obviously a legitimate disambiguation of “the English lan-
guage” on which that expression has not changed its referent in the last twenty-six
years. Let us say that “the English language” refers to the same “languageH” (the sub-
 Chapter 20.  A positive argument against SCT 

script stands for “historical”) as the homonym of that expression twenty-six years ago,
but that “the English language” refers to a different “languageSR” from its homonym
twenty-six years ago (the sub-script stands for “semantic rule”).

An erroneous conception of language

Here are some views that might initially seem reasonable:

Lewis’ term “LANGUAGE” refers to the same thing as Saussure’s term “langue”,
and Lewis’ term “language*” refers to the same thing as Saussure’s term “parole.”
“LANGUAGE” and “langue” both refer to languagesSR; and “language*” and “pa-
role” both refer to languageH.

This is false. The term “linguistic meaning”, I will argue, never refers to function-theoretic
pairings, and the term “language” never refers to a set of such pairings. The statement “a
language is a set of function-theoretic pairings of individual, properties, and so on, with
physical tokens (or types)” is false on every disambiguation of the term “language.”
Lewis is right to say that the word “language” is ambiguous. But it is not ambigu-
ous in the way that he thinks. And Lewis is obviously right to distinguish between se-
mantic and psychological meaning. But he is wrong to think that the former is a pure-
ly function-theoretic relation.
Obviously Saussure is right to distinguish between languages and the use people
make of them. But a language is as much a spatiotemporal entity as the use people
make of it. Langue and parole are both spatiotemporal entities. Not all spatiotemporal
entities are identical with events. Given that an event is a change in spatiotemporal
conditions, it follows that not all spatiotemporal entities are events. A langue is a spa-
tiotemporal condition. Parole is a series of events. But both langue and parole are spa-
tiotemporal entities.
Here we must head off a source of deep confusion. Lewis is, I believe, entirely right
to say that a language is a set of semantic rules. The English language is the set consist-
ing of rules like: “Socrates” refers to Socrates, “Plato” refers to Plato, and so on. The
English language is not a series of noises and ink-marks; it is not even a series of ink-
marks and noises coupled with the socio-psychological conditions that caused them.
When I say “that rabid dog is running towards us”, my utterance is caused by (inter
alia) my awareness of existing semantic rules, the same being true of any other utter-
ance of any language. If utterances constituted language, this would not be possible.
The psychological state that was the proximal cause of my producing that utterance
was itself a causal consequence of the existence of the English language: if it weren’t for
the existence of the English language, I would not (ceteris paribus) have had that men-
tal state; and the dependence referred to here is causal, not logical or analytic or con-
stitutive. Thus, the mental state that precipitated my saying “that dog is running to-
wards us” is not constitutive of the English language: if that mental state were thus
 Conceptual Atomism and the Computational Theory of Mind

constitutive, we couldn’t coherently register the obvious truth that its existence is a
causal consequence of that of the English language. So even though Lewis is (so I be-
lieve and so I will argue) entirely wrong as to what semantic rules are, he is right to
think that a language is a set of such rules.3
Even though the mental state that is the proximal cause of my saying “a rabid dog
is running towards us” is not constitutive of the English language, it doesn’t follow that
no mental states are thus constitutive. Being a non-platonic object, the English lan-
guage is, I will argue, constituted by psychological facts. (It is also constituted by the
social and environmental facts in which these psychological facts are embedded.) But
the psychological constituents of language are distinct from the immediate psycho-
logical antecedents of utterances.
A story may help bring out my meaning. You and I are roommates. We both want to
live in a neat and hygienic environment. If we are to live in such an environment, various
chores must be done on a daily basis – somebody must vacuum, do the dishes, and so on.
We explicitly agree that I will do those chores on even numbered days of the month and
that you will do them on odd-numbered days of the month. Here we have an arrange-
ment whose existence is obviously constituted, in part, by certain psychological entities.
On November 4th, I look at the calendar and realize that it is my chore-day. I thus
bolt into action – I start vacuuming, doing the dishes, and so on. Here the proximal
cause of my doing these chores is my having a certain awareness: I realize that, because
it is an even day of the month, I must do various chores if I am to honor my agreement
with you. But that realization is not constitutive of the agreement that you and I have.
That realization presupposes that agreement, given that the latter is a cause of the
former. So even though our agreement that I do laundry on certain days obviously has
psychological constituents, the immediate psychological causes of my doing laundry
on any given day are not constitutive of that agreement.
We can go further in this direction. Causes are distinct from their effects and caus-
es thus cannot be constitutive of their effects (or vice versa). Given this, suppose we say
that my realizing, on November 4th, that I must do chores is constitutive of our agree-
ment. In that case, we are forced to say that our agreement was in no way a cause of my
doing chores on that day. But that is absurd. The fact that we entered into that agree-
ment is obviously part of what causes me to vacuum, do the dishes, and so on. (It is not
an immediate cause or a total cause; but it is obviously a cause.)

3. Some linguists – so-called “discourse-theorists” – think that languages are just patterns of
behavior and thought. For the reasons just outlined, I believe this to be false. But given only that
discourse-theory is (so I believe) false, and that there are such things as semantic rules, it doesn’t
follow that semantic rules are platonic entities. My guess is that advocates of nihilistic views such
as discourse-theory take it for granted that, if there are semantic rules, they are platonic abs-
tracta. Given this belief, plus the understandable (though, I believe, false) belief that Platonism
is wrong, one has no choice but to deny the very existence of semantic rules. Since, as we are
seeing, semantic rules are not platonic entities, opponents of Platonism do not have to reject the
fruitful and plausible view that semantic rules do exist and guide our thought and conduct.
 Chapter 20.  A positive argument against SCT 

A language is no more a platonic or function-theoretic entity than the agreement

just discussed. Any language is constituted by various understandings that hold among
people. Any language thus has at least some psychological components. But it does not
follow that the immediate psychological causes of one’s utterances are constitutive of
language. Indeed, if we said that the immediate causes of my uttering “that dog is run-
ning towards us” were constitutive of language, then we would be barred from saying
that the existence of a given language was in any way causally responsible for my pro-
ducing that utterance. But surely the fact that the English language is alive (i.e. that it
is not in the same category as Sanskrit or Latin) is a partial cause of my producing that
utterance. Some of my mental states are constitutive of the English language. My
knowledge of certain social and psychological facts is constitutive of the English lan-
guage. (This isn’t to say that there is no possible world where English exists but I do
not. A given molecule can be a constituent of a cloud even though there are possible
worlds where that cloud exists but that molecule does not. Similarly, my having certain
bits of knowledge is constitutive of the English language, even though there are worlds
where English exists and I do not, and even though there will be a time when the Eng-
lish language exists and I do not.) But the psychological states that immediately cause
my utterances are not constituents of the English language.
Parole is not identical with what we referred to as “languageH.” Parole is language-
use, not language-change. Given a set of semantic rules, parole is the use of those rules.
Parole is not the replacement of that set with another such set. Of course, parole is
causally responsible for language-change. It is because of how people use words that
“probably” now means capable of being shown to have a greater than 50% chance of be-
ing true, whereas it used to mean capable of being shown to have a 100% chance of being
true. But it obviously doesn’t follow that parole is constitutive of language change. (in
fact, the opposite follows) “EnglishH” refers to an evolving sequence of sets of semantic
rules – one that includes the semantic rules governing the language that Chaucer spoke
and also governing the language that we speak. Given what we just said, it follows that
use of English expressions is not constitutive of any of the sets including in EnglishH
and a fortiori is not constitutive of EnglishH itself.
It may help if we take a brief detour through metaphysics. What we typically refer
to as “events” are hyper-changes: they are changes in the manner in which things
change. If a woodsman splits a log in two, there has been an event. But it isn’t as though,
prior to that event, the log’s existence involved an absence of change. The log’s exist-
ence supervenes on innumerable changes (e.g. displacements of sub-atomic quanta of
mass-energy). But these changes are all internal to a structure. When the woodsman
splits the log, this structure is altered. So the woodsman’s act is an event not because it
introduces change where there was previously none, but because it altered the frame-
work within which change took place. His event was a change in the manner of change.
Conditions are framework-internal changes; events are framework-external changes,
i.e. they are changes in the manner of change.
 Conceptual Atomism and the Computational Theory of Mind

A set of semantic rules is a structure. A usage of an expression belonging to that

structure is indeed an event, but it is a structure-internal event: it is comparable to the
subatomic mass-energy displacements that constitute the log prior to its being split in
two. A language-change is, of course, identical with or realized by a series of events
that are not structure-internal since, by definition, such events alter the relevant struc-
ture. Here our woodsman-analogy breaks down. The woodsman’s act is entirely exter-
nal to the structure of events constitutive of the log, whereas the events that realize
language change are necessarily on the periphery of the events constitutive of semantic
rules. But what is important in this context is that, so far as events change a language
– so far as they add a semantic rule or change what some expression means – those
events are not constitutive of any existing set of semantic rules, and are therefore not
constitutive of any existing language (on one important disambiguation of that term).
This truth is obscured by the fact, just noted, that the events that constitute language-
change are in some respects (historically, psychologically, semantically) continuous
with those that are internal to a set of semantic rules.

Summary of our discussion thus far

A set of semantic rules is not a set of platonic entities; it is not a set of function-theoretic
assignments of properties, individuals, and so on, to physical tokens. (We haven’t yet dis-
charged this assumption.) A set of semantic rules is constituted by psychological entities of
some kind (understandings that hold among people), and also by the social and environ-
mental underpinnings of those entities.
Of course, the proximal cause of any utterance is a psychological entity. (You see a
rabid dog running towards you. Because of your semantic knowledge, you respond by
saying: “there is a rabid dog running towards me!” The distal cause of your utterance is
an external state of affairs – some dog running towards you. But the proximal cause is
your perception of that state of affairs.) But such proximal causes are never constitutive
of the semantic rules that make up a language.
So even though a language is constituted by psychological entities, and also by
understandings among people that implicate facts about their shared socio-physical
environment, a language is nonetheless not constituted by speech-acts or by the psy-
chological states that proximally cause such acts. Speech-acts, as well as their immedi-
ate psychological antecedents, are effects of the existence of semantic rules, and are
therefore not constitutive of such rules. Since languages are constituted by such rules,
it follows that neither speech acts nor their proximal causes are to any degree constitu-
tive of language.
Lewis’ “language*” is not Saussure’s parole, and Lewis’ “LANGUAGE” is not Saus-
sure’s langue. As Lewis defines them, the terms “language*” and “LANGUAGE” don’t
refer to anything that is actually referred to by the term “language.” Lewis’ terms refer,
respectively, to language-use and to sets of platonic entities. We’ve just discussed why
 Chapter 20.  A positive argument against SCT 

the term “language” never refers to language-use. In the upcoming section, we will see
why “language” never refers to sets of platonic entities.
Language-change is not to be identified with language-use. This fundamental
truth is obscured by the fact that the set of events constitutive of language-change
overlaps to a large degree with the set of events constitutive of language-use.
The word “language” is indeed ambiguous, but not in the way that Lewis thinks.
The expression “the English language” sometimes refers to a set of semantic rules that
are operative at a given time, and it sometimes refers to a series of such sets. In the
sentence “people use language to communicate with one another”, the word “language”
is being used in the first way. In the sentence “languages change over time”, it is being
used in the second way. In the first case, the word “language” refers to a condition that
obtains during a relatively short period of time. In the second case, it refers to a series
of such conditions. In each case, the word “language” refers to a spatiotemporal phe-
nomenon. In neither case, therefore, does it refer to anything platonic.
Henceforth, we will use the term “language” – and “English”, “Spanish”, “Chinese”,
and so forth – to denote single sets of semantic rules, not sequences of such sets.

Why the function-theoretic approach to language is wrong

So far I have asserted without argument that semantic rules are spatiotemporal entities
and are therefore not to be understood as mathematical functions from physical to-
kens to truth-values or properties. It is not hard to supply the needed argument.
Given any two objects, there is some set or ordered pair consisting of them. The
ordered pair <”Plato”, Richard Nixon> exists no less than the ordered pair < “Plato”,
Plato>. But there is no semantic rule that assigns Richard Nixon to “Plato.” Obviously,
the existence of the just mentioned ordered pair is not enough for that. By parity of
reasoning, the semantic rule that assigns Plato to “Plato” cannot be identified with the
ordered pair < “Plato”, Plato>.
There is a rule or function that assigns truth to the sentence “Plato was wise” ex-
actly if Plato was wise. But there is also a function that assigns truth to the sentence
“Plato was wise” exactly if Richard Nixon was wise. There is no English semantic rule
to the effect that “Plato was wise” is true exactly if Richard Nixon was wise. So the ex-
istence of a function of the kind just described is compatible with the non-existence of
a semantic rule assigning truth to “Plato was wise” exactly if Richard Nixon was wise.
In general, the existence of any ordered pair or mathematical function is compatible
with the non-existence of any given semantic rule. Thus, the semantic rule assigning
truth to “Plato was wise” exactly if Plato was wise cannot be identified with a function
that assigns truth to that sentence (or, more exactly, to tokens thereof) exactly if Plato
was wise. Consequently, the semantic rule that assigns the proposition, Plato was wise
to (occurrences of) “Plato was wise” cannot be identified with any ordered pair (with,
 Conceptual Atomism and the Computational Theory of Mind

for example, the ordered pair < “Plato was wise”, Plato was wise> or with any mathe-
matical function.
One might counter-respond by saying:
Given only that the pair <“Plato”, Richard Nixon> exists, it does follow that “Plato”
refers to Richard Nixon, i.e. it follows that it refers to it in some possible but unused
language. And given that there is an abstract function that assigns truth to “Plato was
wise” just in case Richard Nixon was wise, it does follow that “Plato was wise” means
Richard Nixon was wise, i.e. it follows that it means it in some unused language.

The expression “unused language” is a veritable synonym of “a non-existent, but possibly

existent, language”, and this counter-response is therefore purely verbal.
But for the sake of argument, let us suppose that the objector is right; let us suppose
there is some language, albeit an unused one, containing a semantic rule R whereby
“Plato” refers to Richard Nixon. Right now people couldn’t possibly use R. Right now, if
you used “Plato” to refer to Richard Nixon, you would simply be mis-speaking. And this
shows that R is no semantic rule at all.
Of course, we could imagine the English language evolving in such a way that
“Plato” became a name for Richard Nixon. But that evolution would involve creating
conditions that allowed people to refer to Richard Nixon as “Plato.” It wouldn’t consist
in people simply using R. If it exists, R is not available for the using. And if it ever be-
came available for the using, that would not be by virtue of the mere existence of the
ordered pair <”Plato”, Richard Nixon>. (After all, that pair exists right now, even
though R is unusable.) Rather, it would be by virtue of some kind of structural change
on the part of the system of communication used by speakers of English. But in that
case, it would be the resulting structural condition, not the ordered pair, that made it
possible to refer to Richard Nixon as “Plato.” At no point would anyone manage to re-
fer to Nixon by using that ordered pair.
What we just said about that ordered pair is true of any abstract structure – it is
true of an abstract function that assigns truth to FPlato has phiG exactly if Richard
Nixon has phi. The idea that semantic rules are purely function-theoretic entities is
thus a non-starter.

The Gricean approach

Nonetheless, given only that semantic rules cannot be understood in strictly platonic, func-
tion-theoretic terms, it doesn’t follow that the concept of expression-meaning is to be un-
derstood in terms of that of speaker’s meaning. In fact, it is the other way around: the con-
cept of speaker’s meaning must be understood in terms of that of expression-meaning.
As Searle showed (1969), one can affirm that Plato was wise with the sounds “Pla-
to was wise” only if the latter already means the former. In general, one can affirm P by
means of S only if S already means P.
 Chapter 20.  A positive argument against SCT 

Of course, to affirm P, one doesn’t have to use an expression that has P for its lit-
eral meaning. I may convey the proposition your wife is having an affair with the sen-
tence “sometimes things aren’t as they seem.” But that is because, the circumstances
being what they are, there is an existing symbolic relationship – albeit not of literal
meaning – between that sentence and that proposition.
A consequence is that one can try to affirm that Plato was wise with the sound
“Plato was wise” only if one believes that the latter means the former. In other words,
one can mean the proposition Plato was wise with utterance of “Plato was wise” only if
one believes that the latter already means the former. In general, one can mean P with
S only if one believes that S already means P.
To echo a point made a moment ago, it doesn’t follow that, in order to mean P with an
utterance of S, one must believe that S literally means P. I can mean the proposition your
wife is having an affair with the sound “sometimes things aren’t as they seem.” But that is
because I believe that, the circumstances being what they are, there is an existing symbolic
relationship – albeit not of literal meaning – between that sentence and that proposition.
So one cannot affirm P with S unless S already means P, and one cannot even try
to affirm P with S unless one believes that S already means P. Thus, in direct opposition
to what Grice held, the concept of speaker’s meaning is to be understood in terms of
that of expression-meaning.4
Griceans respond to this by saying that expression-meaning is “conventionalized”
or “fossilized” speaker’s meaning. According to this view, Grice’s analysis shows how it
is that “snow is white” came to mean snow is white. But now that “snow is white” has
that meaning, we don’t have to suppose that what it means coincides with what people
mean by it. Its meaning is to be understood not in terms of speaker’s meaning, but
rather in terms of conventions. In their turn, those conventions are to be understood
in terms of speaker’s meaning. This is Simon Blackburn’s view.5
But to say that linguistic meaning is “fossilized” or “conventionalized” speaker’s
meaning is to admit that linguistic meaning is constituted by conventions, and it is
thus to admit that linguistic meaning is ultimately not speaker’s meaning. Thus, if lin-
guistic meaning is conventionalized speaker’s meaning, then Grice’s analysis provides,
at most, a causal or historical explanation as to how it is expressions came to have the
meanings they currently bear. But here we are not looking for a causal-historical expla-
nation as to how E came to have meaning M. We are looking for an analysis of what it
is for expression E to have meaning M, and such an analysis is not to be found in the
supposition that linguistic meaning is fossilized speaker’s meaning.
We have to distinguish the question “what is it for ‘snow is white’ to mean snow is
white?” from the question “how did ‘snow is white’ come to mean snow is white?” The
view that linguistic meaning is conventionalized speaker’s meaning provides no answer
to the first question, and is thus irrelevant as a hypothesis as to the nature of linguistic

4. See Salmon (2005) for a similar argument.

5. See Blackburn (1984: 122-134). Blackburn himself (1984: 113) uses the term “fossilized.”
 Conceptual Atomism and the Computational Theory of Mind

meaning. We may conclude that Grice’s view is wrong. (We will develop this argument
in Chapter 25, and some of the points just made will therefore be repeated.)
But we’ve already seen why, given the erroneousness of Grice’s view, it does not
follow that the function-theoretic view is the right one. There is a third view: semantic
rules are stable psycho-sociological conditions. (In Chapter 25, we will attempt to say
what exactly these conditions are.) We’ve already seen the merits of that view. Thus, a
false dichotomy is involved in the inference, implicit in much modern semantics, from
“Grice was wrong” to “languages are platonic abstracta.”
The analysis just put forth makes it clear that nothing that doesn’t presuppose the
existence of thought bears any significant resemblance to anything that is a semantic
rule of a public language or, therefore, to anything that is uncontroversially a semantic
rule. A language just is a set of semantic rules. So if Mentalese exists, it is a set of se-
mantic rules and its existence therefore presupposes that of thought. This shows that
SCT is indeed guilty of vicious regressiveness.
 chapter 21

Another argument against SCT

The concept of non-conceptual content

There is one problem with SCT that we haven’t yet discussed. In my view, it is the most
significant of the many problems with that doctrine. (In this context, the term “sen-
tence” will cover both sentence-types and sentence-tokens.)
Consider the sentence: “the cat is on the mat.” The meaning of that sentence is
what is referred to as a “proposition.” Presumably, propositions have structures at least
approximately like those of the sentences that express them.1
In fact, this last statement is a tautology. A sentence must belong to this or that
particular language – it must be a sentence of Chinese or English or French. Where
there isn’t a systematic way of assigning propositions to sentences, there is no lan-
guage. Where there is such a system, there is ipso facto a function that establishes an
isomorphism between sentences and proposition. And where there is such a function,
there is ipso facto a resemblance between sentences and propositions.
In any case, it seems likely that the proposition expressed by “Caesar killed Brutus”
has one constituent corresponding to Caesar, another to the relation of killing, and to
a third Brutus. 2 (As we will see later, there is more to that proposition than that – oth-

1. See, for example, Kaplan (1989: 494), R. Moore (1995: 100-115). This view is found in Rus-
sell (1903: Chapter V).
2. There is an obvious challenge to the idea that propositions must have structures similar to
the sentences that express them. “Caesar killed Brutus” and “Brutus was killed by Caesar” pre-
sumably express the same proposition, but those sentences differ structurally. Also, the Alba-
nian translation of “Caesar killed Brutus” is (for all I know) syntactically different from that
English sentence.
This challenge is not hard to deal with. “Caesar killed Brutus” and “Brutus was killed by Cae-
sar” have different derivation-trees. This shows that the way in which Caesar is assigned to the
constituency of what is meant by the one sentence differs from the way in which he is assigned to
the constituency of what is meant by the other sentence. But it doesn’t show that those constituen-
cies themselves differ, either in respect of their membership or their internal organization.
It may be true that the syntax of the Albanian translation of “Caesar killed Brutus” is diffe-
rent from that of its English counterpart. But the discoveries of depth-grammarians show that
these differences are more apparent than actual – that they hold only if the term “syntax” is taken
in a folk-grammatical and pre-scientific sense. In Russian, the present tense of the verb “to be”
is seldom used. One doesn’t say “I am thirsty”, but merely “I thirsty.” On this basis, one might
conclude that the proposition meant by “I am thirsty” cannot resemble the English sentence that
expresses it, given that the corresponding Russian sentence has a different (though, to be sure,
 Conceptual Atomism and the Computational Theory of Mind

erwise it wouldn’t be different from the proposition Caesar killed Brutus or from the
unordered set (Caesar, the relation of killing, Brutus). But we will deal with these is-
sues later in this chapter.)
The fact that sentences decompose into a finite number of discrete parts consti-
tutes a serious problem for SCT. Much information is not digital. Consider your cur-
rent visual perception. It doesn’t decompose into minimal units of significance. Sup-
pose that you actually saw Smith punching Jones. Smith is not a minimal or otherwise
discrete part of the representational content of that perception; and no concept, or
Fregean sense, of Smith is such a constituent. What we just said about Smith is true of

not a very different) structure. Chomskyan linguistics has made a strong case that the Russian
sentence does contain an occurrence of the verb “to be”, i.e. the equivalent of “am”, but that it
contains it “aphonically” (i.e. that occurrence isn’t realized at the level of acoustics of orthogra-
phics). If this is right, then there is no reason to believe that “I am thirsty” is syntactically diffe-
rent from its Russian translation, given only that the latter does not overtly contain an expres-
sion corresponding to “am.”
But even if the argument just given turns out to be fallacious, it is undeniable that sentences
have a certain non-trivial similarity with the propositions they express. Leaving aside the gramma-
tical differences between languages, and between different sentences within a language that ex-
press the same proposition, it is clear that the meanings of sentences decompose into a finite, and
usually very small number of discrete parts. The meaning of “Smith punched Jones” has a discrete
part corresponding to Smith, a discrete part corresponding to “punch” (let us momentarily set
aside the niceties relating to the tense-marker), and a discrete part relating to “Jones.” This pro-
perty will be shared by any sentence having the same meaning – by any translation of that sen-
tence and by any English sentence expressing the same proposition (e.g. “Jones was punched by
Smith”). “Smith punched Jones” has a digital structure: it decomposes into discrete, minimal units of
significance; and its meaning therefore has a structure that decomposes into discrete, mutually
independent parts. What is true of that sentence is true of all sentences of all languages.
It is easy to find pairs of sentences that are synonymous but appear to differ syntactically.
But given any one of these pairs, the line of thought presented a moment ago is easily extended
to show that it is no threat to the notion that propositions have a structure similar to the senten-
ces that express them.
Here is an illustration (again involving Russian). Russian is a highly inflected language. Even
though “Ivan” and “Ivana” are the Russian translations of “John” and “Jane”, one doesn’t translate
“John loves Jane” by saying “Ivan lyubit Ivana” (“lyubit” being the translation of “loves”). One must
say “Ivan lyubit Ivanu” (actually, in Russian, it doesn’t matter, or matters much less than in English,
what order the words have). In Russian, the direct object must be inflected.
Does this mean the proposition meant by “John loves Jane” does not have a structure com-
parable to that of the corresponding sentence? No. As we noted a moment ago, Russian has a
relatively free word-order, whereas English has an extremely rigid word-order (a fact that nati-
ve-speakers of Russian often find to be irritating when translating their thoughts into English).
This is plausibly taken to mean that, in English, word-order does the job that inflections do in
Russian. So, in effect, English does contain inflections (or the equivalent); but whereas inflec-
tions are realized by discrete, phonetically realized entities in Russian, they are realized by word-
order in English.
 Chapter 21.  Another argument against SCT 

Jones and the relation of punching. Indeed, it is true of anything that is the semantic
content of any one of the expressions composing a sentence that describes the state of
affairs in question.
Your perception of Smith involved perceptions of parts of Smith’s body. Indeed, it
involved a sequence of such perceptions. And each of these perceptions involved yet
other perceptions: there are no minimal perceptions. (This makes it questionable
whether there are such things as perceptions. Although there is obviously such a thing
as perceiving, there seems to be no such thing as a perception – just as , although there
is such a thing as watching T.V., there is no such thing as a T.V.-watching. So far as
statements about “perceptions” are meaningful, they are really elliptical for statements
about stretches of perceptual activity. But statements about sentence-tokens are mean-
ingful when taken literally; for there obviously are such things as sentence-tokens. This
line of thought is consistent with the point that we are now in the process of trying to
establish, viz. that analogue representations are not digital representations in disguise.)
Of course, you don’t have an infinitely high-resolution perception of Smith: you don’t
see his microstructure. But even though your perception of Smith does break down, it
doesn’t break down into discrete units of significance. Even if your perception does
have minimal parts – even if, as Hume implausibly maintained, there are minima visi-
bile – these very much seem not to combine in a way at all comparable with the way in
which “Smith”, “Jones”, and “punched” combine into the sentence “Smith punched
Jones.” It is thanks to the operation of recursive rules that components of that sentence
combine into a meaningful whole. But it is hard to believe that similar recursions are
what combine your perception of Smith’s knee with your perception of Smith’s shirt
with your perception of…into a visual perception of Smith punching Jones. I will ar-
gue that such recursions are indeed not operative.
Our sense-perceptions are analogue-representations. The information that forms
the mind-world interface is analogue, not digital. Unless appearances here are radi-
cally deceptive, that information does not have a decomposition into meaningful sig-
nificant units.
Most information is not digital. We will see that digital information is parasitic on
analogue information. The former is a digitization of the latter. The latter is not an
analogicization of the former.
Sentences are, by definition, digital structures. A sentence consists of a finite
number of minimal and mutually discrete units of significance. So far as thought is
mediated by sentences, cognitive content has a digital, not an analogue, structure.
Given this, it is hard to see how a visual perception could consist of a sentence-token-
ing or a set of such tokenings.
At any rate, to make a case that perceptions did have such a structure, one would
either have to use an extremely broad conception of “sentence” – one that, for reasons
similar to those discussed earlier, would trivialize the claim that all ideation is mediated
by linguistic symbols – or one would have to have made some fantastic discovery about
the structure of visual perception: a discovery that could not possibly be made on the
 Conceptual Atomism and the Computational Theory of Mind

basis of a visual perception’s phenomenology, and that would have to be made on strictly
theoretical grounds. But since a visual experience is a bit of phenomenology, it is hard to
see how contrary to what phenomenology tells us any theory could possibly show us that
it had a structure contrary the one it appears to have. At any rate, it is a highly reasonable,
though conceivably defeasible, presumption that perceptions don’t have structures com-
parable to sentences.
It is also hard to believe that all post-perceptual mental content is digital or there-
fore sentential. Of course, some post-perceptual output is sentential. (We say things in
response to what we see, hear, and so on; and such utterances are obviously sentential.)
But much of it is not. We dream, we fantasize, we create paintings: these involve the
generation of analogue, not digital, representations.
Conceivably, one could hold that the contents of our perceptions must be digitized
in order to be stored in memory. If that is right, then the occurrent memories that re-
play our perceptions would result from an analogicization of digital content. But such a
view would be extremely contrived. It is obviously more reasonable to suppose that the
contents of our perceptions are stored in analogue-form, and that the occurrent memo-
ries that replay them do not involve the conversion just described.
Also, our emotional, aesthetic, and ethical reactions seem not to have sentential
structure. They seem to be amorphous and lacking in structure – or, in any case, lacking
in sentential structure. Of course, some have said that aesthetic and emotional reactions
are meaningless – that, in liking Mozart’s music more than Salieri’s, one’s ideation is to
no degree principled and is only causally related to the stimuli that produced it. But
such a view misrepresents our intuitions in the most extreme possible manner.
One could make the following objection to this last argument:
You are confusing aesthetic and moral reactions per se with the phenomenological
expression of such reactions. Obviously the pleasant feeling that Mozart’s music cre-
ates in you doesn’t have a digital structure. But that feeling is merely a phenomeno-
logical expression of an underlying judgment that is not comparably amorphous.

But the phenomenology is surely constitutive of the aesthetic reaction, and is not mere-
ly an effect of it. This is not to say that every affective response you have to a piece of
music is constitutive of a genuinely aesthetic response to it. A piece of music can trig-
ger an emotional reaction that has nothing to do with your estimation of its merits. But
it is hard to believe that all music-induced phenomenology is in this category. It seems
that a being that had no phenomenological reaction to a piece of music (apart from
those that are purely perceptual) would fail to have an aesthetic reaction to it. So it is
not an option to say that one’s aesthetic reaction to a performance of the 21st piano
concerto has a digital structure, given that the phenomenological aspects of one’s reac-
tion obviously don’t have such a structure.
Given that aesthetic reactions strongly appear not to have a sentential structure,
what we must say is not: “aesthetic reactions are completely subjective; no musical
 Chapter 21.  Another argument against SCT 

compositions are better than others.” What we must say is: “awarenesses of objective
facts are not always embodied in sentential or para-sentential structures.”
In my judgment, the nihilistic views about aesthetics and ethics that were advo-
cated by the logical positivists were expressions of the view that anything truth-evalu-
able must be sentence-like. Given how counter-intuitive it is to suppose that no musi-
cal compositions are better than others, or that no acts are morally better or worse than
others, this alone should make us re-think the idea that mental content is sentential or
para-sentential.
It thus seems clear that perceptual content is not sentential and, consequently, that
perceptions are not sentence-tokens and are not at all like sentence-tokens. SCT thus
cannot be true of more than a fraction of our mentation, given how fundamental per-
ception is to our understanding of the external world.
I would now like to provide further reasons to hold that any sentence (regardless of
whether it belongs to an internal language or an external, public language) must be a
distillation of non-digital information. My arguments will be strictly metaphysical in na-
ture and will not rely on the results of experimental or even introspective psychology.

The McDowell-Evans debate

Let us begin with a few points about the history of the problem that we will be discuss-
ing. Gareth Evans (1982: 122–129) deserves credit for making the distinction between
“conceptual and “non-conceptual” content. By “conceptual”, Evans meant “proposi-
tional”; and by “non-conceptual” he meant, “non-propositional.”
Given this distinction, the question immediately arises as to whether we are deal-
ing with two fundamentally different kinds of content or only with two different con-
ceptions or representations of what is, ultimately, some one kind of content. Evans held
that non-conceptual content is not conceptual content in disguise and that it must
therefore be understood on its own terms, and not in terms of the sentences (or prop-
ositions) in which it can, to at least some degree, be embodied.
Evans’ argument for this position is as follows. Perceptions are much more “fine-
grained” than sentences. What you see is never just red. It is always some hyper-spe-
cific shade of red. Since language uses broad rubrics like “red”, it fails to do justice to
the structure of perception. Thus, perceptual content is fundamentally non-linguistic.
McDowell (1994) has a famous counter-argument. We can refer to anything that
we see – e.g. any specific shade of red – with a demonstrative expression (“that exact
shade of red”). In this way, according to McDowell, Evans’ argument is neutralized.
Of course, McDowell is right that one can refer to anything that one sees. But it
doesn’t follow that the structure of a sentence is at all like the structure of a perception.
If anything, the opposite follows. Suppose that I use the expression “that exact shade”
to refer to some specific shade of red that I am seeing. The expression “that shade of
red” obviously doesn’t have a structure comparable to my perception of the red surface
 Conceptual Atomism and the Computational Theory of Mind

in question. The former decomposes into the words “that”, “shade”, and so on. The lat-
ter does not so decompose.
One can invent new words to refer to specific colors (and sounds and fragrances
and…) that one experiences. For example, I might stipulate that “R” is to refer to some
specific shade of red that I am now seeing. Being a simple expression, “R” has no se-
mantic structure. But surely my perception of the red surface in question is not com-
pletely without any articulations.
The spuriousness of McDowell’s argument is seen by giving an argument exactly
analogous to it. Pointing to some elephant, I say “that elephant weighs over a ton.” Here
I have used a demonstrative expression to pick out some specific elephant. But my
perception of the elephant obviously doesn’t have a structure comparable to that of
“that elephant.” Given only that I can refer to specific objects and property-instances
that I see, it doesn’t follow that sentences are representationally similar to perceptions:
it doesn’t follow that perceptions are not fundamentally different from sentences, and
it doesn’t follow that the contents of perceptions are not fundamentally different from
the contents of sentences (propositions).
Also, even though they often refer to what is maximally specific, indexical expres-
sions operate by way of general rubrics. If I point to a red rose and say “that shade of
red is lovely”, my utterance of “that shade of red” refers to some specific shade of red.
But it secures that reference by using the sortal terms “shade” and “red.” So even though
that utterance refers to some specific shade of red, it semantically decomposes into a
part whose semantic content is the concept red, another part whose semantic content
is the concept shade, and so on. But a perception of a shade of red does not have a
comparable decomposition.
In this context, we must take care to distinguish tokens from types. A token of Fthat
phiG may have for its content a specific shade of red. But the content of the correspond-
ing type is given by the rule:
(T) A token of Fthat phiG refers to the contextually salient phi
or, equivalently:
(T*) If, in context C, there is exactly one salient object x having phi, then a token of
F
that phiG in C refers to x. 3
A token of “this shade of red is darker than that shade” may compare two maximally
specific, perceptually present shades. But the rules that assign meaning to such a token
involve only general rubrics, e.g. red and distal object. So, from a semantic viewpoint,
that token is built out of general categories. But the content of your visual perception

3. There are apparent exceptions to my claim that indexicals work through broad rubrics. For
example, one can say “that is lovely” or “that is hot” and be understood. One doesn’t have to say
“that rose is lovely” or “that pan is hot.” But that only means that one can acoustically omit the
semantically present sortal term.
 Chapter 21.  Another argument against SCT 

of a dark red car next to a light red car seems not to decompose into the concepts dark,
light or red. In fact, the content of that perception seems not to decompose into any
concept into which the semantics of a sentence-token used to report that content
would decompose. So even if a sentence-token attributes a specific property to a spe-
cific object, the content of that token decomposes into general categories. Since the
same appears not to be true of sense-experience, we have reason to believe that there
are fundamental differences between linguistic and perceptual representations.
McDowell (1994) has another argument against the viability of the distinction be-
tween conceptual and non-conceptual content. People can say what they see, hear, and
so on. I see Smith punching Jones. I can report this fact: I can say “Smith punched Jones.”
Perceptual content can be verbalized. Therefore the information borne by sentences
must either be identical with, or similar to, that borne by perceptions.4
This argument is really a generalization of the first argument, and it is therefore no
surprise to see that it involves the same non-sequitur. Given only that we can articulate
what we sense-perceive, it doesn’t follow that sentential content is structurally similar
to perceptual content. A curve can be represented as a sequence of line-segments. But
a curve is not a sequence of line-segments.
One might counter-respond by saying the following:
But any finite series of line-segments is, at best, an approximate representation of
a curve; and any non-approximate representation of a curve would have the same
structure as that curve.

But this exposes an even deeper problem with McDowell’s second argument. It very
much seems that any sentence is, at best, an approximate representation of the informa-
tion borne by any sense-perception. Suppose that I see Smith punch Jones, and I then say
“I saw Smith punch Jones.” First of all, I cannot just see Smith. I can see Smith only by
seeing a state of affairs involving him. Similarly, I cannot see any instance of the relation
of punching except by seeing a state of affairs in which such an instance is embedded.
Consequently, if I say correctly “I saw Smith punching Jones”, my statement is an
extremely approximate and course-grained representation of the content of my visual
perception.
Of course, the approximateness of my verbal representation can be lessened. I can
say: “I saw Smith punch Jones; and Smith was wearing a green shirt and he looked
sweaty and upset…” By adding more and more verbiage, I can narrow the gap between
sentential and perceptual content. But given any finitely long verbal description of
what I saw, some content will always be left out (or so our intuitions tell us) – just as,
given any finite number of line-segments, the exact shape of the curve will not be rep-
resented. It thus seems that Evans’ fineness of grain argument ultimately prevails. (Lat-

4. I used to find this argument of McDowell’s compelling. For this reason, in Kuczynski
(2003), I advocated the view that all content was propositional in nature.
 Conceptual Atomism and the Computational Theory of Mind

er I will try to substantiate our intuition that any finitely long sentence will always leave
out some perceptual content.)
McDowell might counter-respond by pointing out that I can use a demonstrative-
expression to refer to the exact content of my visual perception. I can say, for example,
“I will never forget the content of this visual perception.” But this counter-response
wouldn’t support McDowell’s position, since the expression “this visual perception”
doesn’t have a structure remotely comparable to its referent.
As Russell (1903, 1956) pointed out, at least one of the expressions composing a sen-
tence must have a universal for its meaning. Thus, the content of any given sentence is built
up out of general rubrics. If, as appears to be the case, the same isn’t true of perceptual con-
tent, then sentential content is structurally different from perceptual content.

Some other differences between iconic and sentential representations

It is not hard to find support for the view that perceptual content is fundamentally dif-
ferent from sentential content.5 Consider the operations that can be performed on
sentences – disjunction, conditionalization, negation, generalization, and so on. These
operations cannot be performed on maps or on other graphical representations. You
cannot form the disjunction of two maps.
Of course, maps A and B can be disjoined if you stipulate that A and B are equiva-
lent to sentences (or series of sentences) S and S*, and that juxtaposing those two maps
in a certain way tokens the disjunction of S and S*. But in that case, as Rescorla (2003)
points out, A and B are no longer functioning as maps. By the same token, A and B
would cease to function as sentences if you were to consult the internal structure of A
to find out how to drive to Bakersfield.6
But this is only a special case of the more general principle that no recursive op-
eration that can be performed on expressions of any kind can be performed on maps.
A sequence of maps doesn’t compose a sentence. Of course, one could stipulate that an
expression meaning x is larger than y results if a map of x is placed to the right of a map
of y. But such a juxtaposition of maps would not itself be a map and, in that context,
the two maps in question would not be functioning as maps.
These points about maps correspond to points about visual perceptions. Like
maps, visual perceptions are analogue-representations. Two visual perceptions cannot

5. The argument given in this paragraph and the next is clearly stated in an unpublished dis-
sertation by Michael Rescorla (Rescorla, 2003).
In 1995, I provided the same argument in a paper on Wittgenstein’s “picture-theory” of mea-
ning. As I later discovered, that argument is found in much of the literature on the Tractatus.
6. Again, this point is clearly stated in Rescorla (2003), and also in much of the literature on
the Tractatus. In his work in dreams, Freud often makes the same point. See, for example, Freud
(1965: 24).
 Chapter 21.  Another argument against SCT 

be conjoined or disjoined or negated. More generally, none of the recursive operations

that can be performed on expressions can be performed on visual perceptions. This
suggests that the information encoded in visual perceptions is fundamentally different
from the information encoded in sentences.
We can solidify our findings by developing a point that we made in Chapter 19.
There we saw that, contrary to what Fodor alleges, there could in principle be non-
systematic languages. For there is nothing incoherent in the thought that there might
exist a language L satisfying the following conditions. L contains one symbol meaning
snow is white, another meaning grass is green, a third meaning fire is hot, and no sym-
bols besides those three. Those three symbols would have no syntactic structure.
Therefore, no symbol belonging to that language would have such a structure. Such a
language would not be systematic: it would not be capable of expressing any permuta-
tion of any one of the propositions that it can express. For example, L wouldn’t be able
to express the proposition snow is hot or fire is green. (We saw in Chapter 19 that, given
the mere fact that languages like L could exist, Fodor’s second and third arguments for
SCT do not go through.)
But even though Fodor is wrong to hold that, necessarily, all languages are system-
atic, there is a similar point that is not wrong. Given an unsystematic language, the
symbols belonging to it can be assimilated, without any change in their meanings, into
a systematic language. So L could be absorbed into a language L* that is systematic and
that comprises L as a proper part of itself. Suppose that L* comprises connectives cor-
responding to the English “and” and “or”, but is otherwise just like L. In that case, L*
will be able to conjoin and disjoin the sentences belonging to L. And L* will also be
systematic (at least up to a point); for it will be able to express permutations of some of
the propositions that it can express. For example, L* can express either (snow is white
and grass is green) or fire is hot given that it can express (either snow is white or grass is
green) and fire is hot. The important point here is that the sentences belonging to L
don’t have to undergo a change in meaning in order to be inducted into a language,
such as L*, that can combine them into more complex expressions. So even though not
all possible languages are systematic, nonetheless all possible languages can be turned
into systematic ones, without their symbols having to undergo any kind of change in
meaning. To make a similar point: the symbols belonging to an unsystematic language
don’t have to be changed to become capable of being conjoined, disjoined, or other-
wise subject to the recursive operations characteristic of systematic languages.
But, as we’ve seen, analogue representations, such as maps or sense-perceptions, would
have to undergo radical changes in their meanings (if that is even the right word) in order
to be conjoined or disjoined or otherwise subjected to recursive operations. Therefore, at
least some kinds of ideation are not linguistic or even similar to anything linguistic.
Incidentally, some would say that, if the symbols of L were inducted into a lan-
guage that had sentence-level operators, those symbols would ipso facto change their
meanings. So-called “meaning-holists” have such a view. (See, for example, Wittgen-
stein 1958, Field 1977, Brandom 1998.)
 Conceptual Atomism and the Computational Theory of Mind

But meaning-holism is false. During the last twenty years a number of new terms
have been added to English, e.g. “cyberspace” and “e-mail.” But such additions have not
changed the meaning of “snow” or any other English term.
Meaning-holism involves a failure to distinguish meaning-giving from reference-
fixing. Suppose that, tomorrow, it becomes common knowledge that Aristotle (not Plato)
wrote The Republic. In that case, when children ask “who was Aristotle”, we may say:
(*) “Aristotle was the author of The Republic.”

So it seems as though a term’s meaning changes when we learn new facts about its
referent. For similar reasons, the creation of terms such as “e-mail” may change how
we define old terms. For example, we might now define the term “pen pal” thus:
(**) “a pen pal is somebody to whom one routinely sends, and from whom one rou-
tinely receives, amicable letters or e-mails.”

But the meaning of (*) is:

(*WS) Somebody x uniquely wrote The Republic, and “Aristotle” names x.
(The “WS” stands for “wide-scope.”)
And the meaning of (**) is:
(**WS) Given any medium of epistolary communication M (be it snail mail or e-mail or
telegraph...), the term “pen pal” is to be defined thus: x and y are pen pals just in
case they exchange friendly communications by means of M.

In each of (*WS) and (**WS), the definition proper is confined to the underlined part.
The rest merely helps identify the right definition. So while facts about reference-fixing
are affected in very obvious ways by changes in our knowledge-base, and by additions
to our lexicon, the same is not true of facts about meaning-giving. Meaning-holism
wrongly takes facts about reference-fixing to be facts about meaning-giving. (See Fo-
dor and Lepore 2002 for a similar argument.)
 chapter 22

Propositional structure and the ineliminability

of non-conceptual content

I would like to develop the arguments given in the previous chapter. Consider the
sentence:
(*) “Smith has red hair.”
It is said that “Smith” denotes an “individual.” Notice that, etymologically, “individual”
means undividable. But from a metaphysical and also a representational point of view,
so-called individuals, like Smith and Jones, are anything but undividable. (If, like Plato,
you believe that souls are indivisible, then assume that “Smith” and “Jones” refer to
rocks or plants.) And yet from a linguistic point of view they are undividable: “Smith”
and “Jones” are minimal, undividable units of significance.
In this section, I want to show that SCT cannot be right, the reason being that
language necessarily represents as fundamental what is extremely derivative from the
viewpoint of how we mentally represent the world. Even though, linguistically, “Smith”
is a minimal unit of significance, Smith himself is best thought of, not as an “undivid-
able” or even as a thing at all, but rather as a property of sequences of sets of properties
of properties. (More accurately, Smith is an instance of a property of a sequence of sets
of instances of properties of properties.) And even though “red” is semantically basic,
neither the property of redness nor any instance thereof is in any sense fundamental. In-
stances of properties, I will argue, are abstract commonalities holding among situa-
tions; they are not components of situations. The world isn’t built up out of instances of
properties, and our perceptions of the world are not out representations of such instances.
(Your perception of the red square doesn’t consist in a perception of an instance being
red conjoined with a perception of squareness.) But any sentence necessarily decom-
poses into units that represent these highly derivative entities (if “entity” is even the
right word: see below) as basic.

Properties versus hyper-properties

First of all, the term “red” doesn’t denote anything of which there could possibly be a
spatiotemporal instance. No fire-truck or rose or pool of liquid is just red: it is some
specific shade of red. (There are no general terms indicating the exact shade of any
 Conceptual Atomism and the Computational Theory of Mind

specific red object. One must use demonstrative expressions to indicate any maxi-
mally specific shade of red – any kind of redness that can actually be seen.)
What we generally refer to as “properties” are actually properties of properties.
Suppose that R1 is the specific shade of red had by object x1 (some fire-truck), that R2
is the specific shade had by x2 (some rose), and so on. R1 is something of which there
could be, and are, instances in the spatiotemporal world; the same is true of R2, R3, and
so on. By contrast, the thing denoted by “red” is something that R1…Rn have in com-
mon. So “red” denotes a property of properties, not a property simpliciter.
Perhaps this argument demands elucidation. There is some property that all and
only those objects instantiating R1 have in common, the same being true of Ri, for ar-
bitrary i. Thus, so far as there is some property that all and only so-called “red” objects
have in common, that is by virtue of there being some property that all and only R1…
Rn have in common. x1 (the fire-truck) and x2 (the rose) don’t have quite the same
color. Strictly speaking, they have different colors.
In any case, it is a datum that x1 and x2 differ chromatically. And it is a datum that
the difference between them is comparable to, though much smaller than, the differ-
ence between a blue and a green object. The difference between x1 and x2 is to the
chromatic difference between a diamond and an emerald what the difference between
3.1 and 3.2 is to the difference between 2 and 978,965. We are dealing with different
degrees of some one kind of difference. So x1 and x2 have different colors. (Anyone who
described x1 and x2 as having the same color would be speaking either loosely or false-
ly.) So far as x1 and x2 are similar in respect of color, it is because R1 and R2 have some-
thing in common. Thus, so far as they both fall under the rubric “red”, it is because
their colors have a common property. So “red” denotes a property had in common by
different colors. Redness thus turns out to be a property of properties.1

1. Here we must head off an objection to the line of thought just presented:
Your analysis confuses the concept of an instance with that of a sub-category. (Equivalently,
it confuses the concept of a “determinable” with the concept of a “determinate”, to use the medieval
terms.) R2 is not an instance of the concept of redness. Rather, it is a sub-category of that property.
Let r2 be the specific instance of redness associated with the rose. r2 is not, whereas R2 is, something
of which there can be instances. r2 is an instance both of R2 and of redness. Redness is a property not
of R2, but of r2. R2 is a sub-category of the category (or property) of redness, not an instance of it.
And, contrary to what you say and to what is patently obvious, r2 does instantiate redness no less
than it instantiates R2. So redness is not a property of properties. It is just a property.
A property is something of which there could be multiple instances. It is a “way that things
are”, to use David Armstrong’s (1989) expression. In any case, it is a way that things could be.
(There is a subtle and, in this context, irrelevant qualification to this platitude, relating to the
supposed existence of properties that cannot possibly be had and that, consequently, cannot
correspond to ways that things might be. An example would be the property picked out by the
predicate “round square.” But, I would suggest, that predicate is best thought of as expressing a
concept that fails to pick out a property. So rather than saying that “round square” picks out a
property that nothing could have, we should say that, although its grammatical function is to
 Chapter 22.  Propositional structure and the ineliminability of non-conceptual content 

What we just said about “red” is true mutatis mutandis of any predicate that one
would find in any natural language. So far as there exist natural-language expressions
that denote properties of spatiotemporal entities, as opposed to properties of such
properties, it is in virtue of the occurrence of some demonstrative expression (“that
exact shade of red”) or some artificial extension of a language (such as our introduc-
tion of the predicates “R1”, “R2”, and so on).
Neologisms such as “R1” must be defined by means of demonstrative expressions
(e.g. “that color”). As we saw in the last chapter, even though a token of “that exact
color” picks out R1, or some other specific color, it secures that reference by way of
general rubrics (e.g. color and distal object). So the content of an expression either de-
composes into general rubrics or it must be understood in terms of such rubrics. For
reasons that we’ve already considered, this suggests that sentential content is funda-
mentally different from perceptual content and, therefore, that the contents of our
perceptions are not propositions.

pick out a property, it fails to do so – just as we say that “the rational square root of two” fails to
pick anything out, even though its grammatical function is to do so.) So far as any spatiotempo-
ral thing x is red, it is because x has R1 or R2 or…There is thus a clear-cut and non-circuitous
sense in which R1 is a way that spatiotemporal things can be.
But there is only a circuitous sense in which redness is a way that such things can be. After
all, a thing can be red only by way of being R1 or R2 or...This confirms the supposition that red-
ness is, in some fundamental respect, not comparable to R1 or R2 or any other property that is
clearly had by individuals. That supposition is further confirmed by the fact that there is a non-
circuitous sense in which “red” describes a way that a property could be. There is, no doubt, some
property R that each of R1…Rn has and that specific shades of blue and specific temperatures do
not have. (When I say “specific shades of blue and specific temperatures”, I am referring to pro-
perties and not to instances of properties; I am referring to things of which there could be ins-
tances, and not to such instances themselves.) Of course, R is nothing other than the property of
redness. Supposing that B3 is some specific shade of blue, there is some property that B3 does not
have that any one of R1…Rn does have. There is, to use Armstrong’s language, a way that any one
of R1…Rn is that B3 is not.
To sum up, it seems a datum that “red” denotes something that different specific colors have
in common. Given this, it follows that redness is a property of colors, and is thus a property of
properties. The latter properties have spatiotemporal instances. But we mustn’t say that redness
itself has such instances. If we say that a given rose instantiates both R2 and some property that
R2 itself has, then we end up embracing the view that a property of an abstract object can also be
a property of a spatiotemporal object. But this would blur the line between the particular and the
universal – between the spatiotemporal and the non-spatiotemporal. And if anything is certain
in metaphysics, it is that this line must be kept sharp.
There are indeed expressions that denote sub-categories, as opposed to instances, of the
thing denoted by “red.” Examples are “burgundy”, “maroon”, and so on. But two things that are
burgundy can differ in respect of color. By contrast, two things that are R2 cannot so differ. Thus,
predicates like “burgundy” and “maroon” are not comparable to predicates like “R1” and “R2.”
 Conceptual Atomism and the Computational Theory of Mind

Why property-instances are abstract entities

Let r1 be a specific instance of R1. r1 is not a constituent of the situation in which it is found.
There will never be an instance of color without an instance of shape or an instance of
location. Language separates out what are, in actuality, completely inseparable. The color
of the rose cannot be detached from its shape or its location.
Situations are not assemblages of property-instances. On the table, there is an ob-
ject that is square and green. Let S be the situation just described. S is not composed of
an instance of greenness (or of some specific shade thereof), and an instance of square-
ness, and an instance of being on the table. When we verbally describe the situation, we
use different terms to denote these different property-instances. But the composition
of the sentence isn’t comparable to the composition of the situation.
There couldn’t possibly be an instance of squareness that wasn’t an instance of a
certain size and an instance of having a certain location. Although we use sentences to
describe situations, situations are not structured in the same way as sentences. The
question thus arises as to how sentences can be even approximately accurate represen-
tations of situations. Ian Hacking (1975) refers to this as the “articulation problem.”
What we just said about situations is true mutatis mutandis of our perceptions of
them. Seeing that there is a box of a specific color and shape on a specific table does not
consist in producing mental representations of multiple property-instances and then
assembling them. Your seeing the box on the table does not, and could not, consist in
your having a mental representation of some specific shade of green, and also of some
specific shape, and then somehow putting these representations together.
Of course, you don’t have to see the box as green. You might see it as red. But what
we just said (mutatis mutandis) would apply to this situation: your seeing the box
wouldn’t consist in your combining a mental representation of an instance of square-
ness with a representation of an instance of redness. Perceptions do not decompose
into representations of instances of properties. (In fact, perceptions don’t really “de-
compose” at all. The very concept of composition seems inapplicable here.)

The consequences of these points for SCT

But if we thought in a language, your seeing the box would consist in exactly such an
act of assemblage. Any sentential representation of S will separate what cannot possi-
bly be separated in S itself or in any perceptual representation thereof. A sentential
description of S will inevitably have one expression denoting one property-instance, a
separate expression denoting a separate-property instance, a third expression denoting
the table, and so on. It makes no difference, in this context, what kinds of properties
the expressions in question denote. They can denote bona fide instances of spatiotem-
poral objects (e.g. they can denote specific colors and specific shapes); or they can de-
 Chapter 22.  Propositional structure and the ineliminability of non-conceptual content 

note properties had in common by such specific properties (“green”, “rectangular”).

Either way, what will be fused in the situation, and in any perceptual representation
thereof, will be separated in the sentence.
An obvious consequence is that a linguistic representation of the situation consists
in cobbling together distinct expressions that denote what are, in the situation and also
in a perception of the situation, inseparable entities (for lack of a better word). So sup-
posing that mental representation consisted in the tokening of sentences, seeing S
would consist in your (at some level) assembling a visual representation out of an ex-
pression denoting a specific color, and a separate expression denoting a specific shape,
and so on. But it is absurd to suppose that this is how it works.

The concept of a property-instance

Since instances of properties are not detachable parts of situations, they are better
thought of as “dimensions of similarity.” For two situations to instantiate some property
is for them to be similar in some respect. Their both instantiating that property is to be
understood in terms of their being similar in some respect – not the other way around.
Obviously expressions like “instance of squareness” and “instance of R1” (where R1
is some maximally specific shade of red) are meaningful. But given any instance of any
property – given anything that can be described as an “instance of R1” – that instance
cannot be distinguished, except in a purely notional sense, from various instances of
what are obviously different properties. Given some specific instance of R1, that in-
stance cannot be distinguished, except via a process of abstraction, from a certain
shape and from a certain location. So to the extent that it denotes an isolable entity, a
token of “that instance of R1”, or of “that instance of red”, denotes an abstraction, and
does not denote a constituent of the spatiotemporal world. (What is such a constituent
is the situation in which that instance is embedded.)
According to Russell (1905, 1918), a significant noun phrase that (so to speak)
denotes a purely notional entity is one that doesn’t denote anything, and that must be
understood contextually. “The average person” is obviously a significant expression.
We could say that it denotes a purely “notional entity” or an “idealization.” But this
would just be a dodge. When you say that “the average person needs more than four
hours of sleep a night”, you are not saying that an idealization needs more than four
hours of sleep a night: abstract objects don’t sleep. There is indeed no x such that you
are saying of x that x needs more than four hours of sleep.
Thus, “the average person” isn’t significant by virtue of picking some object out. It
is significant by virtue of the fact that there is a rule assigning meaning to sentences of
the form F…the average person…G 2

2. Russell held that definite descriptions in general are in the same category as “the average
man.” For reasons given earlier, I believe that he is wrong about this. But this has no bearing on
 Conceptual Atomism and the Computational Theory of Mind

If, as we’ve argued, expressions like “instance of red” and “instance of R1” denote pure-
ly notional entities, then they don’t denote anything and must be understood contextually.
Another example may give some weight to this line of thought. The expression “the
temperature of this object” is obviously significant, and that is why it can be a part of mean-
ingful utterances, e.g.:
(*) “the temperature of this object is greater than the temperature of that object.”
But no occurrence of “the temperature of this object” denotes an entity. (*) doesn’t say
that one thing is greater than some other thing. The definite descriptions in (*) are to
be understood contextually. And when we produce a sentence that perspicuously rep-
resents the proposition affirmed by (*), the result is a sentence that says how two things
compare in some respect. (When you say that x is hotter than y, you are saying that the
number that results when the net kinetic energy of x’s constituent molecules is divided
by the number of such molecules is greater than the number that results when the net
kinetic energy of y’s constituent molecules is divided by the number of such molecules.
The term “temperature” has parsed out, and we are left with a statement that shows
only how x and y compare in some very abstract respect.) I would suggest that what is
obviously true of the expression “the temperature of this object” is true of all expres-
sions that, so to speak, refer to instances of properties. Such expressions don’t refer to
anything; and the propositions affirmed by sentences containing them concern how
one situation compares, in some respect, with some other situation.3
Anything that is uncontroversially a sentence contains at least one expression re-
ferring to a property. So if, as advocates of SCT hold, we think in sentence-tokens, then
any (truth-evaluable) mental representation consists of a sequence or juxtaposition of

his view that the significance of non-denotative noun-phrases is to be explained, not by ontolo-
gizing, but by identifying the relevant contextual definitions.
3. There is a caveat here. When Russell said that “the king of France” must be defined contex-
tually, he meant that any given sentence of the form Fthe king of France has phiG is synonymous,
or at least analytically equivalent, with some sentence not containing any apparent reference to
a French monarch. Leaving aside the empirical knowledge in which one’s knowledge of English
semantic rules is embodied, no empirical knowledge is needed to know how “the king of France”
parses out of “the king of France is bald.” (Russell didn’t arrive at this Theory of Descriptions by
doing experimental psychology. He did so only by reflecting on his existing knowledge of En-
glish semantics.)
But when I say that “the rod’s having a temperature of 78°” is to be understood contex-
tually, I don’t mean that it is analytic, or (what is sometimes different) a mere matter semantics,
how it is to parse out. I am saying only that it ultimately meets the same fate as “the average man”
and “the limit of F(x) as x approaches 2.” Any occurrence of “the rod’s having a temperature of
78°” conceals a great deal of relational structure. An obvious corollary is that, so far as a sen-
tence doesn’t conceal that structure, it doesn’t contain an occurrence of that, or any comparable,
expression. If we are right, then what we said about that particular definite description is true in
general of expressions referring (so to speak) to property-instances.
 Chapter 22.  Propositional structure and the ineliminability of non-conceptual content 

discrete entities, at least one of which denotes an instance of a property. Thus, my see-
ing the green round thing on the table consists in (inter alia) some word-token denot-
ing an instance of a certain color being juxtaposed with a word-token denoting an in-
stance of a certain shape. But such a view doesn’t seem to be true to the actual
composition of my perception.

Expressions denoting individuals

What we said about property-instances, and about our mental representations thereof, ap-
plies with extra force to so-called individuals. Let us once again consider our paradigm:
(*) “Smith has red hair.”
Semantically, “Smith” (the expression, not the person) is a simple entity, the same be-
ing true of any proper noun. But from the viewpoint of metaphysics and also of per-
ceptual representation, Smith is not a simple entity. In fact, it isn’t even clear if, from
those viewpoints, he is an entity at all.4 (Once again, if you believe that people are in-
divisible, or otherwise metaphysically special, then suppose that “Smith” refers to a
rock or a carrot.)
During any given interval of time, Smith’s existence supervenes on the occurrence
of innumerable events – psychological, metabolic, cellular, molecular, atomic, and ul-
timately sub-atomic. At any given instant, i.e. during any arbitrarily small period,
Smith’s existence is realized by a set of states of affairs. (Here the word “set” is not being
used in its strict mathematical sense.) Thus Smith’s existence as a whole, i.e. from his
birth to his death, is realized by a sequence of sets of states of affairs. Smith is therefore
anything but a simple entity.
Given any event E involving Smith, E’s occurrence supervenes on those of innu-
merable lower-level states of affairs. More precisely, E occurs in virtue of the fact that
some causal sequence, consisting of a series of sets of contemporaneous states of af-
fairs, acquired some property that it didn’t previously have. So if it is true, “Smith sud-
denly realized that 7+5=12” holds in virtue of the fact that some pattern of mass-ener-
gy displacements took a new turn. In terms of metaphysics (not semantics), the
occurrence of “Smith” in that sentence parses out.
Smith is only notionally separable from the various processes that constitute him. For
the reasons given a moment ago, it follows that occurrences of “Smith” ultimately parse
out – like occurrences of “the rod’s having a temperature of 78°” and “the average man.”
We saw earlier why Smith must mentally be represented by a description (or, more
accurately, by one’s knowledge that some description is uniquely satisfied), even though
(if direct-reference theorists are right, as I believe they are) “Smith” is semantically

4. See Merrick (2001) for a different argument for a viewpoint not unlike that about to be
presented.
 Conceptual Atomism and the Computational Theory of Mind

simple. So, in thought, Smith is not represented as a simple. This point of ours, as well
as our arguments for it, may not meet with universal acceptance. (Fodor 1998 says that
my concept of Smith has no representational complexity.) But what cannot reasonably
be debated is that, in perception, Smith is represented only to the extent that some situ-
ation in which he is embedded is represented. Thus, considered as an isolable entity,
Smith is a non-entity from the viewpoint of sense-perception.

Revisiting the distinction between data and meta-data

This line of thought relates to a point that we defended in Chapter 4. When someone
says that he saw Smith, he is making a judgment about what was given to him in a
visual perception; he is not giving a mere report as to what was given to him.5 What
was actually given to the percipient in sense-perception is some piece of existential
information that Smith satisfies.
Let us discuss the nature of the judgment just referred to. (Doing this will involve
some repetition of points made in Chapter 1.) Smith has an identical twin, Twin-Smith,
who is now in Finland. You are now in Delaware, and so is Smith. You know all of this,
and you are now seeing Smith. As we discussed, given only the information encoded in
your visual perception, you could just as well be seeing Twin-Smith. When you say
(correctly) “I saw Smith, not his twin”, you are not merely reporting what you saw: you
are making a judgment about it, and that judgment incorporates your knowledge that
Twin-Smith is in Finland. What is given to you in sense-perception is therefore not
Smith is wearing a green shirt but (so far as it can verbalized) is more along the lines of:
there is exactly one thing x such that x has such and such properties (e.g. the property of
standing in such and such a place at such and such a time); moreover, x is wearing a green
shirt. When you judge that you saw Smith, you are judging that Smith is the one who
satisfies that description, i.e. that he is the right value of the variable.
But what is it for Smith to satisfy this description? Smith’s satisfying that description
consists in its being the case that the state of affairs that is given to you in your perception
has a certain continuity with other states of affairs; and Twin-Smith’s not satisfying that
description consists in its not being the case that the state of affairs that is given to you in
your perception has that same kind of continuity with those states of affairs.
Smith is not hidden behind a veil consisting of some state of affairs. Given only that
you must judge that you are seeing Smith, and that this fact cannot be encoded in strictly
perceptual information, it doesn’t follow that your senses are imperfect or that sense-
perception is a flawed medium. What follows is a fundamental metaphysical fact about
Smith. Smith’s being the person in the green shirt consists in its being the case that there
is some state of affairs, involving the wearing of a certain green shirt, that has a certain
continuity with certain other states of affairs.

5. See Russell (1927: 40) for a similar point.

 Chapter 22.  Propositional structure and the ineliminability of non-conceptual content 

Suppose that your sense of sight is maximally good – that you see as well as it is
nomically possible for any being to see. And suppose that you witness every moment
of Smith’s life from his birth to his death. For the reasons given a moment ago, your
saying “I saw Smith today” on Monday will not be a case of your simply reporting what
was given to you in your Monday-perception. Even though you see everything that
there is to be seen, the information encoded in your Monday-perception will be given
by an existence-claim that, in principle, could be satisfied by somebody other than
Smith. Even under these idealized circumstances, when you say on Tuesday “I saw
Smith on Monday”, your statement reflects your knowledge that the state of affairs that
you saw on the previous day has a certain continuity with various other states of affairs
(that you have either seen of otherwise sence-perceived, or that you know about indi-
rectly, viz. through inference or testimony). Suppose that you confined yourself only to
the information encoded in the visual perception that you had on Monday, merely
reporting its contents, and making no meta-perceptual judgments – even ones that,
given your high degree of knowledge, would be extremely conservative. You still
wouldn’t use the word “Smith” or any equivalent. Your reportage would be given by an
existence-claim that described a situation but didn’t contain any mention of Smith;
and Smith’s satisfying that existence-claim would consist in the just mentioned situa-
tion’s standing in a certain relation (of causality or continuity) to other situations.6

6. One point must be made clear. Let us continue to suppose that you have maximally acute vi-
sion and a maximally good memory. (In this context, any mention of vision is to taken as a refe-
rence to any kind of sense-perception.) Under these circumstances, you would, of course, be right
to say “I saw Smith yesterday”; and, as we’ve just emphasized, that statement would embody a
meta-perceptual judgment. But it doesn’t follow that your statement would involve any element of
conjecture. By hypothesis, you have seen (and, we may suppose, remembered) everything that
there is to see. (Of course, Hume argued that any use of perception or memory involves an element
of conjecture – and perhaps he was right. In that case, my point is that, leaving aside the element
of conjecture inevitably involved in any use of perception or memory, your statement that you saw
Smith does not express a conjecture, even though it does express a judgment.)
When a physicist conjectures that a certain kind of particle-jump is responsible for a certain
macroscopic effect, he is going beyond what he sees. He sees a needle-displacement. He infers
that a certain kind of micro-event occurred. He is making a judgment and a conjecture.
When you (under the idealized circumstances described a moment ago) say “I saw Smith
yesterday”, your statement does indeed involve your going beyond what is given to you in the
visual perception that you had yesterday. So relative to that particular perception, your state-
ment is not a case of innocent reportage. But there is nothing (in this context) that you haven’t
seen; and your statement is therefore not a conjecture. You don’t have to make a leap into the
unseen. You must relate what was given to you in your yesterday-perception with what was gi-
ven to you in perceptions from earlier days; and doing this would obviously involve certain co-
gnitive abilities: it would presuppose the ability to generate memories of perceptions and to or-
ganize these memories in a certain way. But your statement that you saw Smith would not reflect
your having made a theoretical leap.
 Conceptual Atomism and the Computational Theory of Mind

With a few exceptions, what we just said about Smith is true of practically any-
thing. It isn’t true of abstract entities. Maybe it isn’t true of instantaneous, sub-atomic
mass-energy displacements (supposing that it even makes sense to speak of “instanta-
neous” displacements). But it is true of tables, vases, people, and cats. It is true of any
spatiotemporal entity that lasts for any amount of time (and is thus, it seems to follow,
true of everything spatiotemporal).
Let us now discuss the relevance of this to SCT. Every human language contains
proper nouns (e.g. “Smith”) and noun-phrases (e.g. “the guy over there wearing the
green shirt”) that can substitute for proper nouns when the latter are not available.
Nothing that is at all like a public language would lack expressions of the kind just
discussed. If a representational medium doesn’t contain nouns (or noun-phrases) that
refer to enduring entities like Smith, then that representational medium will be very
different from anything that is a clear case of a language.

Why no language could possibly mediate perception or cognition

Earlier we briefly discussed some reasons why any language must contain expressions de-
noting broad rubrics, as opposed to perceptible entities. Here we will develop those points.
Suppose that there is some proper noun N whose semantic content is some spe-
cific situation. Two people who have not both personally witnessed that situation can-
not use that noun to communicate with each other – unless, of course, they have some
way of indicating what is meant by that noun that does not presuppose their both hav-
ing witnessed that situation. But any such second method of communication would it-
self involve some symbol (or set of symbols) S and what we just said about N would be
true of S: if S’s semantic content is some object or state of affairs that not both people

Now let us consider the case where you are subject to human limitations – where your vi-
sion and memory are not maximally powerful. Under these circumstances, when you say “I saw
Smith yesterday”, your statement would involve an element of conjecture. But the conjecture
wouldn’t involve your positing some special entity. The conjecture would be to the effect that,
mediating between the situation that you saw yesterday and other situations that you saw on
previous occasions (or, at least, that you knew about on the basis of other situations that you had
seen), there were other, comparable situations that you didn’t see. The physicist’s conjecture in-
volves the positing of something very different from what he has seen or even could see. When
the physicist conjectures that a certain micro-event is responsible for a certain perceptible dis-
turbance, he is not merely positing that there exists a certain continuity between certain seen (or
seeable) situations. (A hundred years ago, philosophers of science would have said that this is
precisely what such a conjecture consists in. But it has been reasonably well established that this
is false.) But when you say that you saw Smith, the conjecture involved in your statement (so far
as any conjecture is involved) is precisely that, mediating between the situation that you saw
yesterday and other situations that you saw on prior occasion, are various other comparable, but
(by you) unseen, situations.
 Chapter 22.  Propositional structure and the ineliminability of non-conceptual content 

have personally experienced, then S’s usefulness as a means of communication between

those two people involves some third symbol (or set of symbols) S* in terms of which
S’s semantic content can be described. And, of course, what we just said about S is true
mutatis mutandis of S*.
Sooner or later, we must arrive at a system of communication that does not pre-
suppose experiences of numerically identical situations. We therefore need a system of
communication whose building blocks refer to what is common to different situations:
a system whose basic units refer, not to concrete entities, but to abstract commonalities
holding among concrete states of affairs.
Given a system of communication of the kind just described, there is no difficulty
creating ways of referring to specific situations, even to specific psychological events –
events that are not, and cannot possibly be, experienced by distinct individuals (even
though, of course, different individuals can experience qualitatively identical such
events). But reference to such situations will be achieved through symbols that indicate
commonalities among distinct situations. I can refer to a specific experience by saying
“the pain I am feeling right now, in this part of my foot”; and this expression is meaning-
ful and intelligible to others. But the expression through which I refer to this experience
is constructed out of expressions that don’t refer to anything specific and that have for
their meanings abstract commonalities holding among different states of affairs.
The more a system of communication depends on its users having shared experi-
ences, the less able that system is to say what is not already known. And the more ab-
stract are the significations of the expressions composing a system of symbolism, the
less that system depends on its users having shared experiences. This means that ceteris
paribus the more powerful a system of communication is, i.e. the more able it is to
impart what isn’t already known, the more abstract the significations of its most basic
expressions must be. Ceteris paribus the more powerful a language is, the less its basic
expressions are connected with anything that is found in the spatiotemporal world or,
therefore, with anything that can be perceived or otherwise experienced.7
Thus, languages necessarily invert the structure of thought. Given the concept of what
a language is, and of what it is for one language to be expressively more powerful than
another, it is not an option to say that thought is mediated by a language or indeed by any-
thing that is not as unlike a language as a representational medium could possibly be.
What we just said about interpersonal languages is true of intrapersonal languages, sup-
posing that the latter exist. Any instance of communication involves transmitting informa-
tion to a place (so to speak) where it was previously absent. This may involve the transmis-
sion of information from one mind to another or, arguably, from one part of a given mind to

7. Except in so far as there are “experiences” of platonic abstracta. But while I have no doubt
that we can grasp platonic abstracta – that, indeed, we routinely grasp them quite directly, i.e.
without the mediation of symbols – it seems false to say that we experience them. It seems, in
fact, what we mean by the word “experience” is a kind of awareness that has concrete, not abs-
tract, entities for its objects (or that, like a tickle or itch, has no object at all).
 Conceptual Atomism and the Computational Theory of Mind

another part of that same mind. (For the record, although I believe that Fodor is wrong to say
that thought categorically consists in the tokening of expressions of an internal language, I do
not think that thought categorically does not consist in such tokenings. It seems quite possible
that bona fide systems of symbolism are involved in communication between different cogni-
tive spheres - even though it seems less plausible, and indeed may be tautologously false, to
suppose that symbols mediate information-exchanges within such a sphere. So I believe it to
be an empirical question whether there are thought-languages, even though it is not an em-
pirical question whether all thought involves language, the reason being that, as we’ve seen, it
is incoherent, and therefore analytically false, to suppose that all thought is linguistic.) So far
as a language’s basic expressions denote concrete situations, that language is intelligible to
different entities only to the extent that those entities have had experiences of numerically
identical entities. But so far as the latter condition is met, the need for communication plum-
mets, as does the possibility of it, given that, by hypothesis, the sender and the recipient are
both in possession of the same empirical data.
So to the extent that communication between different parts of a mind is necessary,
the communiqués in question would be constructed out of expressions that have highly
abstract entities for their semantic contents. This means that they would have for their
contents things that couldn’t possibly be perceived or otherwise experienced. So wheth-
er we are dealing with inter- or intra-personal communication, language inverts the
structure of cognitive content. (At any rate, it inverts the structure of one very funda-
mental kind of cognitive content: the content of experience. Presumably it is true of
some other kind of cognitive content: the kind that one has in virtue of being a linguistic
creature.) SCT is therefore inconsistent with basic facts about the nature of language.

Individuals as properties of worlds

In this section I will argue that what we refer to as “individuals” – people, rocks, trees,
vases – are better thought of as properties of worlds. This will confirm the points al-
ready made about the viability of SCT. It will also provide a basis for the positive an-
swer that we will give (in the next chapter) to the question “what are propositions and
how, if it all, are they different from the information encoded in our perceptions?”
Suppose that S1…Sn are the various mass-energy displacements occurring in the
space-time region occupied by Smith. In that case, any possible world that comprises
S1…Sn comprises Smith. So Smith’s existence is realized by or supervenes on the occur-
rence of S1…Sn. At the same time, we don’t want to say that Smith is identical with any
such sequence; for presumably Smith could have had experiences and done things
other than those that he in fact had and did.
Nonetheless, Smith is not modally infinitely elastic. There are surely some limits as
to how, and how much, he might have differed. Smith didn’t have to become a lawyer;
he could have become an actor instead. But he couldn’t have been a rock or a carrot.
 Chapter 22.  Propositional structure and the ineliminability of non-conceptual content 

Less trivially, if Kripke’s (1980: 110–115) plausible views on the essentiality of ori-
gins are correct, Smith couldn’t have had parents different from those that he actually
had. More generally, in any possible world where he exists, Smith’s beginnings must be
much the same as they are here. So supposing that s1…sm are the events that, in actual-
ity, initiated Smith’s existence, any possible world where Smith exists is one that com-
prises either s1…sm or events that are significantly like them and, indeed, that at least
partly coincide with them. Thus, Smith exists in a world W exactly if certain events or
states of affairs hold in W – exactly if, to use Kenneth Taylor’s (1998) suggestive expres-
sion, W’s “quantum” is “rippled” in a certain way.
To drive this point home, I am going to use a metaphor that I will soon re-use in
my efforts (in Chapter 24) to give a precise answer to the question “what are proposi-
tions and how, if it all, are they different from the information encoded in our percep-
tions?” Suppose that God wanted to make Smith exist in some world W. He would
have to wrench W’s quantum in the right way; He would have to make sure that certain
mass-energy displacements occurred. If Smith doens’t exist in a world W* whose low-
est-level occurrences are e1…en, it is not an option – even for God – to make Smith
exist in a world W whose lowest-level occurrences are qualitatively just like e1…en. Just
as a metal-worker must dent a sheet of metal in a certain way if he is to ensure that it
will be an auto part (of a certain kind), so God must dent W’s quantum in a certain way
if He is to ensure that Smith exists in that world.
Here we must remember what we said about property-instances. Yesterday I saw a green
shirt, and I thus saw an instance of green. But I didn’t just see an instance of green. I also saw
an instance of a certain shape, and of a certain shape, and so on. So even though I saw an
instance of green yesterday, what I really saw was a green situation; and in saying that “I saw
an instance of green”, I am abstracting from what was actually perceptually given to me.
Similarly, in saying that I saw Smith, I am abstracting greatly from what I actually
saw. What I actually saw was obviously not just Smith, but was rather some complex
situation. So far as that situation involved Smith, it was not because Smith was an in-
gredient of it, but rather because that situation had the right relations with other situ-
ations (relations of both of causation and, to a lesser degree, of resemblance). So far as
he is an isolable entity, Smith is a feature, not a constituent, of situations.
Thus, so far as “Smith”, or any other equivalent term, is a basic constituent of a
language, that language represents as basic what is in fact an extremely abstract prop-
erty of situations. Such a language represents as basic what can only be grasped by
abstracting from data that is itself abstracted from perceptual experience – data that,
indeed, must be abstracted from abstractions of data that, in its turn, is abstracted from
perceptual experience.
The only things that are not abstract – the only things that are seen or touched – are
situations. Situations are not articulated in a way that corresponds at all closely to verbal
descriptions of them. (In fact, it isn’t clear whether they are “articulated” at all: the con-
cept of articulation seems inapplicable. For it seems to me that articulatedness is a
property of digital representations: to be articulated is to be digitized in a certain way.
 Conceptual Atomism and the Computational Theory of Mind

In Kuczynski 2006 I discuss how, in light of this point, Hume’s arguments concerning
induction and causation prove to be spurious and how it also becomes possible to find
a principled basis for inductive and abductive inference.) The rubrics used by language
– “Smith”, “pain”, “red” – do not correspond to anything that could possibly be encoun-
tered in a visual or otherwise sensory experience. For this reason, any medium that is at
all language-like inverts the structure of cognitive content – except in the extremely
rare cases where that content has been distilled, through a process of analytic reflection,
into sentential or parasentential representations – an immediate consequence being
that SCT is false.
 chapter 23

Conceptual content and the structure

of the proposition

As we noted, Evans held that there is “non-conceptual” content. Others have disa-
greed. We’ve seen some reasons to think that Evans is right. But to arrive at a definitive
resolution of this dispute, we must make it maximally explicit what is meant by the
term “non-conceptual content.”
It is a truism that sentences have meaning. (In this Chapter, “sentence” is to be taken
to mean “sentence-token”, wherever appropriate.) Further, it is a truism – a matter of ter-
minology – that these meanings are propositions. The question “is there non-conceptual
content?” seems to amount to this: “Are there representational entities whose contents are
not propositions and are not, in any significant respect, even proposition-like?”
We’ve already seen that the answer to this question is: “yes.” But, in the present au-
thor’s view, no unexceptionable answer to the question “what are propositions?” has ever
been produced. And so long as that question remains unanswered, the opponent of non-
conceptual content can take refuge in the obscurity of the notion of propositionality.
I believe that, given some of the points made in the last chapter, we can develop a
creditable analysis of what propositions are and, on that basis, can further undermine
the credibility of the view that there is no non-conceptual content.

What are propositions? The traditional answers

Let us briefly discuss some of the existing answers to the question “what are proposi-
tions?” First, there is Russell’s (1903) answer. The proposition meant by “Smith punched
Jones” is some kind of a structure that consists of Smith, the relation of punching, and
Jones. (Let us set aside the problems relating to the tense-marker.)
The problem with this answer is that, while it may be right, it is excessively vague. We
want to know what exactly the just-discussed structure is. The most obvious answer to this
last question is: “that proposition is the set (Smith, the relation of punching, Jones).”
I do not think that, historically, anyone has given this exact answer. But it is worth
considering because the problems with the answers that have been given can be under-
stood in terms of the problems with this answer.
First of all, if the proposition meant by “Smith punched Jones” is the set (Smith,
the relation of punching, Jones), then the proposition meant by “Jones punched Smith”
must also be that set. In that case, those propositions would be identical. Since they are
 Conceptual Atomism and the Computational Theory of Mind

not, the proposition meant by “Smith punched Jones” is not that set. In general, the
problem with the answer being discussed is that it cannot distinguish between distinct
propositions that have the same constituents.
Many (probably most) contemporary authors do hold that the proposition meant
by “Smith punched Jones” is a set. They deal with the problem just discussed by saying
that it is a structured set (R. Moore 1995: 101, Cresswell 1985: 446–452, King 1995,
1996, 1997, Perry 2000: 213). So, according to most proponents of this view, Smith
punched Jones is the ordered pair <<Smith, Jones> the relation of punching>, whereas
Jones punched Smith is the ordered pair << Jones, Smith> the relation of punching>. In
general, propositions are structured sets; and distinct propositions that have the same
constituencies are different structurings of those constituencies.
There are two problems with this position. The proposition Smith punched Jones is true
or false. But the set <<Smith, Jones> the relation of punching> is neither true nor false.1
This problem is not necessarily insuperable: if we identify some relation R such
that <<Smith, Jones> the relation of punching> has R with respect to all and only those
worlds where Smith punched Jones is true, then we can identify <<Smith, Jones> the
relation of punching> with that proposition, and we can identify its being true (or
false) in a given world w with its bearing (or failing to bear) R to w. (Of course, care
must be taken that our choice of R doesn’t render this analysis of propositionality triv-
ial or circular.) But advocates of this view have identified no satisfactory value of R
(even though, as we will see, some unsatisfactory attempts to do so have been made).
Consequently, this analysis of propositionality remains programmatic.
There is another problem. For the sake of argument, suppose that Smith punched
Jones is in fact identical with some structure consisting of Smith, Jones, and the rela-
tion of punching. The structure in question cannot consist of Smith’s standing in the
relation of punching with respect to Jones. For if the proposition Smith punched Jones
were identical with a structure of that kind, then the very existence of that proposition
would demand its truth. But that proposition obviously could be false. So if we identify
that proposition with a structure consisting of Smith, Jones, and the relation of punch-
ing, we must make sure that it does not consist of Smith’s standing in the relation of
punching with respect to Jones.
But if the way in which Smith, Jones, and the relation of punching are arranged in
Smith punched Jones is so different from the way in which those three things would be

1. I recall John Perry making this point. But I haven’t been able to find the article where he
makes it.
 Chapter 23.  Conceptual content and the structure of the proposition 

arranged in an instance of Smith’s punching Jones, then it becomes unclear why that
structure is the right one.2
Presumably, the arrangement in question is an abstract, not a physical, one. But that
raises the question discussed a moment ago. Given some particular abstract relationship,
why is that relationship so special? If that abstract relationship has no obvious counter-
part in a case of Smith’s punching Jones, then why say that Smith punched Jones is a
structure that involves interrelating Smith, Jones, and punching in that particular way?
In any case, until we are told exactly what that structure is, the view in question (namely
that Smith punched Jones is a structure consisting of Smith, Jones, and the relation of
punching) is an analysis-schema, not an analysis proper.
It seems that two issues are being muddled by those who identify propositions
with structures of this kind. We can represent the structure of the sentence “Smith
punched Jones” in different ways; for example, we can use a tree-diagram or we can use
an expression like “<<Smith, Jones> the relation of punching>.” (Again, we are setting
aside niceties relating to the tense-marker.) In fact, linguists often do represent senten-
tial-structure in this way.
But this doesn’t tell us anything that we didn’t already know concerning the struc-
ture of that proposition. As many, most famously Wittgenstein (1922), have pointed
out, given that sentences express propositions, there must be some structural similar-
ity between the sentence and the proposition. But until the exact similarity-relation is
identified, we only have an innocuous truism, not an analysis.
This brings us to a related point. The ordered pair <7,5> isn’t true or false. Given
that ordered pairs (and n-tuples) are not in general true or false, why do they suddenly
acquire this property when their members are Smith and Jones, as opposed to num-
bers? Obviously the ordinal properties of <<Smith, Jones> the relation of punching>
are meant to indicate some kind of structure – some kind of structure not had by Jones
punched Smith. But, as we said a moment ago, given only that Smith punched Jones can
be coded in that ordered pair, we know nothing that we didn’t already know as to the
identity of that structure; we know nothing more than we already knew about it, given
our pre-existing knowledge that it can be coded in the sentence “Smith punched Jones”
or in a tree-diagram of a certain kind.

2. This refutes the Tractarian view that the proposition Smith punched Jones is some structure
(consisting of Smith, Jones, and the relation of punching) that is isomorphic with the kind of state
of affairs that would make that proposition true. If that structure is an isomorph of such a situa-
tion, then it would be such a situation (unless the word “isomorph” is being used in a strictly me-
taphorical, and thus completely uninformative, manner). But in that case, the mere existence of
the proposition Smith hit Jones would pre-determine its truth. Given any contingent proposition
P, an analogue of the argument just given shows that, if the Tractarian view is correct, P’s existence
demands its truth. Since some contingent propositions are false, so is the Tractarian view.
 Conceptual Atomism and the Computational Theory of Mind

Frege’s view

Frege’s (1891, 1892) view regarding the proposition Smith punched Jones is that it is a
structure of some kind or other. (He said very little as to what exactly that structure
would be.) But Frege denied that it is Smith and Jones per se that are constituents of that
structure, holding instead that it is concepts of Smith and Jones that are such constitu-
ents. As we discussed in earlier chapters, Frege held that “Smith” and “Jones” have senses
for their semantic contents, where a “sense” of an expression denoting x is a concept that
x uniquely satisfies. Thus, in Frege’s view, it is these senses, and not Smith and Jones per
se, that are constituents of the proposition meant by “Smith punched Jones”, i.e. of the
proposition Smith punched Jones.
In Chapter 3, we saw why, if the semantic content of “Smith” is a concept, then the
proposition meant by “Smith punches Jones” becomes an existence-claim, as opposed
to the atomic proposition that it obviously is. This is not to mention that, given Kripke’s
(1972) points, it isn’t really an option to say that the semantic content of “Smith” is a
concept. From this viewpoint, Frege’s view is a step backwards from the view that Smith
punches Jones is an ordered n-tuple of the kind just discussed.
Frege made a point that could be interpreted as being an attempt at a partial answer
to the question “what is the proposition Smith punched Jones?” Frege said (correctly, no
doubt) that this proposition has as a constituent the concept x punched y. In his view,
the proposition Smith punched Jones is what results when the variables (or empty-spac-
es) are “saturated” with Smith and Jones (or, rather, with the corresponding concepts).
There are two problems here. First, as Davidson (1967) said, until we are told what
“saturated” means, Frege has given us only a worthless metaphor, given that the word
“saturate” is obviously being used metaphorically in this context. Frege has told us that
a proposition is what results when that concept is completed in some way or other by
(senses of) Smith and Jones. But he hasn’t told us what the nature of that completion
is. And until we know that, we don’t know anything that we didn’t already know.
There is another problem. What is the concept x punched y? Sometimes Frege an-
swers this by saying that it is what is had in common by all and only Smith punched
Jones, Mary punched Fred, Larry punched Harry, and so on. So concepts are abstract
features of propositions; they are what different propositions have in common.
But, in that in case, propositions cannot be composed of concepts. (There is, of
course, an obvious parallelism between this point and the point we made earlier con-
cerning the constituencies, or lack thereof, of concrete situations Given that property-
instances are what situations have in common (or, more accurately, are to be under-
stood in terms of what situations have in common), situations cannot be composed of
such instances.) So if we see Smith punched Jones as having the concept of punching as
a constituent, then we cannot see that concept as being abstracted out of propositions.
In general, the constituents of propositions cannot themselves be understood in
terms of propositions. We cannot say that Smith (or redness or pain or…) is what is had
 Chapter 23.  Conceptual content and the structure of the proposition 

in common by all and only those propositions that (in English) are expressed by sen-
tences of the form F…Smith…G (or F…red…G or F…pain…G or…).
In its turn, this fact gives us yet another reason to hold that the contents of our men-
tal states are ultimately not proposition-like and are therefore not sentential. To grasp
Smith punched Jones, I must grasp its constituents. Trivially, I grasp those constituents by
way of mental states whose contents are propositions or by way of mental states whose
contents are not propositions. In the latter case, there is non-propositional content.
In the former case, we have a vicious regress. Grasping a proposition involves
grasping its constituents, and grasping any one of its constituents involves grasping
one or more other propositions. Given the reasonable supposition that no proposition
can be a proper constituent of itself, it follows that one must grasp infinitely many
propositions to grasp a single one. But it is implausible, and (I believe) incoherent, to
suppose that one’s grasping Smith punched Jones involves one’s grasping infinitely
propositions.
Also, given that one cannot grasp a proposition without grasping its constituents, it
is viciously circular to suppose that mental representations categorically have proposi-
tions for their contents. It must therefore be assumed that, ultimately, the contents of our
mental representations of propositional constituents are non-propositional. (Of course,
the contents of such representations are sometimes propositional, given that some propo-
sitions have other propositions as constituents. Hence the word “ultimately” in the sec-
ond to last sentence.) We have seen that there is independent support for this position.

Propositions as the sets of their logical consequences

According to Carnap (1937) and Clarence Lewis (1946), a proposition is the class of its
logical consequences. This thesis is often expressed in terms of sentences: “the content
of a sentence is the class of its analytic [or logical] consequences.” 3
There are some serious problems with this view. First of all, the statement “propo-
sitions are classes that contain their logical consequences” is viciously circular. Given
the reasonable supposition that those consequences are themselves propositions, the
position in question amounts to the circular claim that a proposition is the set of prop-
ositions that follows from it.

3. Carnap (1937: 42) writes: “by the logical content of [a sentence]…we understand the class
of non-analytic consequences of [that sentence].”
Clarence Lewis (1946: 55) writes: “The intension of a proposition comprises whatever the
proposition entails: and it includes nothing else” (quoted in Bonjour 1998: 40).
This view is not an anachronism. Brandom (1998: 96) holds it, and it also underlies the
contemporary view that propositions are sets of worlds (or, equivalently, functions from worlds
to truth-values).
 Conceptual Atomism and the Computational Theory of Mind

One way to block this vicious circularity would be to say that the consequences of
a proposition are not propositions. Whatever merits this position has, it is obviously
not available to someone who denies that there is any fundamental difference between
propositional and non-propositional content.
There is another problem with the view that a proposition is the set of its own
consequences. Presumably the proposition Smith punched Jones has one constituent
corresponding to Smith, one corresponding to Jones, and one corresponding to the
relation of punching. (This is not to deny that it has other constituents. But it presum-
ably has at least those three.) A related point is that the sentence “Smith punched
Jones” surely has a decomposition not wholly unlike that of the corresponding propo-
sition. But it hard to see how the class of logical consequences of Smith punched Jones
could have a structure at all like that of that sentence or that proposition.
This fact is important from the standpoint of somebody who wants his theories to
explain, or at least be consistent with, the fact that languages can be learned and with the
related fact that propositions can be grasped. As Fodor (1975, 1998) makes clear, anyone
who understands “Smith punched Jones” can also understand “Jones punched Smith.”4
Anyone who can understand “Mary is tall”, and “Fred is short” can also understand
“Fred is tall” and “Mary is short.” These facts cannot be explained unless it is assumed
that the things meant by “Fred is tall” and “Fred is short” have a constituent in common
corresponding to the word “Fred” – unless, to generalize this point, it is assumed that the
things meant by sentences typically have discrete constituents corresponding to the ex-
pressions composing those sentences. The class of logical consequences of “Fred is tall”
(or of the proposition Fred is tall) does not have any discrete part corresponding to
“Fred.” In any case, it would take considerable artifice to find any such part.

Why these points are consistent with our analysis of properties and individuals

The points just made might seem in tension with the analysis given in the last chapter.
There we argued that terms that expressions denoting individuals, e.g. “Fred” and “Sal-
ly”, are to be understood contextually: they are not to be understood by pairing them
off with the right entities, since such entities are not to be found. But in the last para-
graph we argued that the thing meant by “Fred is tall” must have a constituent corre-
sponding to Fred.
Given an analysis due to Richard Montague (1974), this tension is easily elimi-
nated. According to Russell, if an expression parses out, i.e. if it is to be understood
contextually, then it has no “meaning in isolation” and therefore fails to correspond to

4. Of course, a person could know what is meant by “Jones punched Smith” without knowing
what is meant by “Smith punched Jones.” An example of such a person would be someone who
doesn’t speak English but has been told what “Smith punched Jones” means. But, as we discussed
in Chapter 13, such a person doesn’t really understand that sentence.
 Chapter 23.  Conceptual content and the structure of the proposition 

a discrete entity. Given this apparently reasonable view, it does seem that we are guilty
of inconsistency. After all, we said that the propositions meant by “Fred is tall” and
“Fred is short” have some part in common corresponding to the word “Fred”, and we
also said that (at the level of metaphysics, not of literal meaning) “Fred” parses out. But
if Montague is right, Russell’s analysis is fallacious, the reason being that it conflates
two totally different notions: the notion of having a referent, on the one hand, and the
notion of having an isolable meaning, on the other.
Consider the sentence “something snores.” For the reasons that Russell and Frege
gave us, “something” doesn’t refer to some ambiguous object. (If, for some x, “some-
thing snores” said that x snored, then “something snores, and something does not
snore” would be necessarily false, as it would affirm the singular proposition x both
does, and does not, snore. But that sentence affirms an empirical truth.) On this basis,
Russell said that it has “no meaning in isolation” and must therefore be defined contex-
tually. But given only that an expression must be defined contextually, it doesn’t follow
that it has “no meaning in isolation.”5 We can see the expression “something” as having
for its semantic content a function that assigns truth to a property (or concept) phi
exactly if phi is instantiated.
We don’t want to say that “something” refers to such a function, given that “some-
thing snores” doesn’t say of some function that it snores. But not all forms of significa-
tion are identical with reference.
It was Frege (1892b) who gave a cogent argument for this last point. Consider an
arbitrary sentence, e.g. “Smith punched Jones.” One might be tempted to say that
“punched” refers to the relation of punching. But if we replace the occurrence of
“punched” with an expression that uncontroversially does refer to that relation, e.g. “the
relation of punching”, the result is “Smith, the relation of punching, Jones.” Since that
string of words is not a sentence, it follows that the occurrence of “punched” in “Smith
punched Jones” does not – at least not merely – refer to the relation of punching. So
unless one takes the demonstrably false view that meaning is categorically identical
with reference, there is no barrier to saying that “something” has for its meaning a dis-
crete entity, notwithstanding that “something” cannot be defined denotatively and must
therefore be defined contextually.
In general, given only that some expression “parses out”, it doesn’t follow that it
doesn’t have a discrete signification. In fact, as I show in Literal meaning and Cognitive
Content (Kuczynski 2005c), given any expression E that parses out, the method that
Montague uses in connection with quantifiers (e.g. “something”) is easily adapted to
show that E has a function for its meaning.

5. As we saw, all expressions – even “Socrates” and “Plato” – are to be defined contextually. To
say that “Socrates” refers to x is to say that expressions of the form FSocrates has phi G are true iff
x has phi. To point to an object x and say “that is Socrates” is to say, implicitly, that an utterance
of the form FSocrates has phiG is true iff x has phi. So denotative definition collapses into contex-
tual definition.
 Conceptual Atomism and the Computational Theory of Mind

Our point in the previous chapter was not that Smith doesn’t exist or that there
aren’t instances of pain. Our point was not that there is no isolable entity of any kind
corresponding to the word “Smith” or to “this instance of pain.” It is that no isolable
constituents of the spatiotemporal world correspond to such expressions. The things
meant by those expressions are more abstract than is usually thought. As we saw, any
instance of pain, even an instance of some maximally specific kind of pain, is really a
commonality holding among states of affairs. In its turn, that maximally specific kind
of pain is a commonality holding among such commonalities. The property pain is a
commonality holding among such commonalities holding among such commonali-
ties. Similar remarks are true of “Smith.”
This shows that “Smith” and “this instance of pain” don’t have isolable spatiotem-
poral significances. But it doesn’t show that they don’t have isolable significances tout
court. They do have such significances, these being the commonalities just discussed.
So given only our views as to what Smith is, and as to what properties and property-
instances are, we are by no means barred from holding that “Smith” and “this instance
of pain” have isolable significances. Our view was that those significances are not spa-
tiotemporally discrete entities – and that view is an independently plausible one.

Propositions as sets of their logical consequences (continued)

We’ve already discussed two difficulties for the view that the proposition Smith punched
Jones is identical with the class of its logical consequences. There is another well known
problem. Smith punched Jones has the same logical consequences as Smith punched
Jones and 1+1=2, even though those are distinct propositions.
Given this fact, it is not an option to say that Smith punched Jones is identical with
the class of its logical consequences – for there is no one such class. Thus if we were to
hold onto the view that propositions are identical with classes containing their logical
consequences, we would have to say that Smith punched Jones was identical with some
structuring or ordering of those consequences and that Smith punched Jones and 1+1=2
was identical with a different ordering of those same consequences. In general, we
would have to say that propositions were orderings of their own logical consequences,
and that analytically equivalent propositions were different orderings of the same
propositions. Otherwise, equivalent propositions would be no more distinguishable
than the unordered sets (2,4,6) and (4,6,2).
But here there arises an analogue of a problem already discussed. It isn’t clear what
the right structuring would be, and the view just discussed is entirely programmatic
until that structuring is identified. So the view in question is, from that viewpoint,
comparable to the view that Smith punched Jones is <<Smith, Jones>, punched>. The
former view, like the latter, is correct only given some as of yet unknown fact about the
structures of propositions. But knowledge of the identity of that structure is precisely
what we want from an analysis of propositions.
 Chapter 23.  Conceptual content and the structure of the proposition 

Also, any set S that contains all the logical consequences of a given proposition P
contains P itself, and also contains either P or there are square circles, and so on. So, on
pain of vicious circularity, we cannot identify P with any set that contains all of P’s
logical consequences. At most we can identify P with some subset of those conse-
quences. But then the question arises what the right subset is.
I do not believe that this particular problem is fatal to the view that propositions
are sets of their own consequences; for I believe that it is not hard to identify the rele-
vant restrictions. But this isn’t a matter that is worth pursuing since we’ve seen that,
independently of this issue, the view in question is either wrong or programmatic to the
point of emptiness.

The possible-worlds approach

Possible Worlds Semantics (PWS) is the thesis that a proposition is a function from
worlds to truth-values. So Smith punched Jones is a function that assigns truth to worlds
where Smith punched Jones and falsity to worlds where Smith didn’t punch Jones (but
where Smith and Jones exist), and that assigns either falsity or no truth-value to worlds
where either Smith or Jones doesn’t exist.
Any adequate assessment of the merits and demerits of PWS would be quite an in-
volved task.6 Fortunately, given our purposes, only a few brief remarks are necessary. If
by a “world” is meant a set of propositions, then it is viciously circular (and regressive)
to say that propositions are functions from worlds to truth-values. If we say that worlds
are non-propositional representations, then we are conceding that there is non-proposi-
tional content – this being exactly the point that we are trying to establish in this chapter.
If we say that worlds are concrete entities, like our world, then PWS involves the dubious
view that, for each possible proposition, there is an actual world where it is true.
PWS also has the problem that it cannot distinguish between 1+1=2 and triangles
have three sides, given that those propositions are true in the same possible worlds.
Advocates of PWS deal with this by saying that propositions are structured assign-
ments of truth-values to worlds (Lewis 1975, Cresswell 1985: 446–452). So even though
1+1=2 and triangles have three sides assign the same truth-values to the same worlds,
they do so in different ways. In one case, the concept of addition is involved in the as-
signment; in the other case it is not. In one case, the concept of triangularity is involved
in that assignment; in the other case it is not. Those propositions thus have different
internal structures and are therefore distinct.
Far from saving PWS, this position eviscerates it. For, according to it, propositions
are individuated entirely by their internal structures: the worlds to which they assign
truth-values drop out, and the heart of PWS is thus abandoned.

6. In Kuczynski (2007), I try to give a relatively complete account of what the problems are
with PWS, and also how PWS could try to circumvent them.
 Conceptual Atomism and the Computational Theory of Mind

Permit me to clarify this last point. We have no idea what Little Timmy’s belief is if we
know only that his belief is true in every world. For given only that information, that belief
could be any one of the following:
(i) 1+1=2
(ii) there are continuous functions that cannot be differentiated at any point.
(iii) There are infinitely many primes.
(iv) Arithmetic is incomplete.
(v) Triangles have three sides.
If we want to know what Timmy believes, we also need to know the identities of the con-
stituents of the proposition believed in, and we need to know how those constituents are
arranged in that proposition. But given that information, there is no need to know in what
worlds that belief is true, since we already have all the information we need to answer the
question “what is the identity of the proposition that Timmy believes to be true?”
Of course, given an answer to that question, we know in which worlds that propo-
sition is true and false (i.e. to which worlds it assigns truth and falsity). In other words,
we know what is sometimes referred to as the “extension” of that proposition. But, as
we’ve just seen, a proposition’s extension is, by itself, useless in the way of indicating
that proposition’s identity, since infinitely many distinct propositions will have the
same extension.
So if it is to avoid falsely identifying 1+1=2 and triangles have three sides, PWS
must see propositions as being individuated by their structures. But in that case, the
extensions of those propositions drop out; and PWS ends up being a version of the
theory, already considered, that a proposition is a structure of some kind. PWS thus
ends up collapsing into (a close relative of) the view that Smith punched Jones is the
ordered pair <<Smith, Jones>, punched>.
There is yet another problem. Suppose that there is only one concrete world. In
that case, snow is green and coal is purple assign the same truth-values to the same
worlds, since they both assign falsity to this world and this world is the only one. So if
those propositions are to be given different extensions, we must embrace the dubious
view that there are, indeed, different concrete worlds. (We’ve already discussed why it
is not an option, at least not in this context, to identify worlds with either sets of prop-
ositions or non-propositional representations.)7

7. This is only the beginning of a statement as to what is wrong with PWS. In Kuczynski
(2006), I provide a more detailed assessment of PWS. But for our purposes, the remarks made
in the preceding paragraph ought to be enough to show that, in this context, PWS does not
provide a viable answer to the question “what is a proposition?” I leave it open whether there are
contexts, e.g. contexts of a purely logical or mathematical kind, where PWS is adequate.
 Chapter 23.  Conceptual content and the structure of the proposition 

David Lewis’ analysis: propositions as properties of worlds

According to David Lewis (1986), a proposition is a “property of a world.” Let P be the

proposition Smith punched Jones. That proposition is true in some possible worlds and
false in others. And for that proposition to be true in a world W is for W to have cer-
tain properties; it is, to use Taylor’s expression, for W’s quantum to be rippled in a
certain way. Hence Lewis’ view.
Permit me to clarify the viewpoint just presented. An artist cannot just project an
image of Lincoln onto a canvass. He must do so one drop of paint at a time. (Even if an
artist could instantaneously project Lincoln onto a canvass, that would consist, not in
his not having to put the requisite thousands of paint drops on the canvass, but in his
being able to put them all there simultaneously.) Similarly, given some contingent
proposition P, even God cannot just make P be true in W. He must do so by wrenching
W’s quantum into shape – by guaranteeing the occurrence of the right micro-events.
God cannot make P be true in a world W whose micro-constituents are qualitatively
identical with those of a world W* where P is not true. Even He must comply with
metaphysical necessities. As Wittgenstein (1922) said, even God cannot create a world
where one and one don’t make two.
In wrenching W’s quantum into shape, what is God doing? He is making a dent in it
– he is doing to it what the artist does to a canvass or what a metal-worker does to a sheet
of aluminum. He is making it instantiate the right property. So for P to be true in W is for
W to instantiate a certain property. (The property in question is not the trivial one of be-
ing a thing x such that P is true in x. Rather, it is the property of comprising micro-events
that collectively form a certain pattern. It is the property, in other worlds, of having a lot
of little ripples that form a big ripple of a certain kind. And since each of those little rip-
ples can be understood independently of P, Lewis’ analysis isn’t trivial or circular.) So P is
a property of worlds, and P is true in a world if that world instantiates it. What we said
about P is true mutatis mutandis of any proposition. Therefore propositions are proper-
ties of worlds.

Evaluating Lewis’ analysis

There is much truth in this analysis, and it is dramatically better than the others we
have considered. In fact, the analysis that I will propose draws heavily from it. But
there are two reasons why Lewis’ analysis is unacceptable as it stands.
If I say to you “Smith punched Jones and 1+1=2”, I am not telling you anything
about how this world differs from others that I am not telling you by saying either
“Smith punched Jones” or “Smith punched Jones and triangles have three sides.” (At
the same time, in uttering any given one of those sentences, I am giving you informa-
tion – though not spatiotemporal information – that I would not be giving you in ut-
 Conceptual Atomism and the Computational Theory of Mind

tering either of the others.) So if Smith punched Jones is a property of worlds, i.e. if it is
a way that certain worlds are, that property is not different from Smith punched Jones
and triangles have three sides or Smith punched Jones and 1+1=2.
Incidentally, Lewis (1986) is aware of this, and he bites the bullet. He takes the
logical positivist view that the three propositions just mentioned are actually
identical. But it is probably best to assume with most philosophers that such a view is
not correct, and to proceed accordingly.
Some abbreviations will help us state a related problem with Lewis’ analysis. Let P1 be
the proposition Smith punched Jones; let P2 be the proposition Smith punched Jones and
1+1=2; and let P3 be the proposition Smith punched Jones and triangles have three sides.
No two of P1, P2, and P3 have the same decomposition. But, as we just saw, given
any one of those propositions, there is no property that a world W has in virtue of that
proposition’s being true in W that W doesn’t have in virtue of either of the other two
proposition’s being true in that world. So if they are properties of worlds, any two of P1,
P2 and P3 are identical and thus have identical decompositions. So Lewis’ analysis can-
not accommodate the fact that distinct propositions can be analytically equivalent.
Nor can it accommodate the related fact that any proposition has a unique decomposi-
tion into a finite number of discrete parts.

Our analysis: propositions as sets of properties, truth as instantiatedness

We’ve considered a number of answers to the question “what is a proposition?” and

found them to be inadequate. I would thus like to propose a different answer.
The core of our view is present in Lewis’ analysis: propositions are properties, and
they are true when instantiated. Truth is instantiatedness. Actually, we will see that
propositions are sets of properties, not properties per se, and that for a proposition to
be true is for a set of properties to be instantiated. But there is obviously a deep kinship
between our proposal and Lewis’.
There is a plant next to my desk. Let D be that plant. D is green. Hence the propo-
sition:
(*) D is green
 Chapter 23.  Conceptual content and the structure of the proposition 

is true. What is involved in (*)’s being true? It is true in virtue of the fact that various prop-
erties are instantiated. The property of being a certain kind of object (a plant) is instantiated
in a certain place and time, and the property of being green is instantiated by that object.8
If various properties are instantiated in a certain place, at a certain time, then (*)
is true. If those properties are not instantiated in that place, at that time, then (*) is
false. This suggests that (*) is a set of properties, and that for (*) to be true is for the
members of that set to be instantiated.
The problem is to identify a set that has the right ordinal or structural properties. More
exactly, the problem is to identify a set that satisfies two requirements. First, (*) is true ex-
actly if all the members of that set are instantiated. Second, facts about the membership of
that set can be put into an intuitively satisfying correspondence with facts about the decom-
positions of sentences like “D is green” (and “that plant over there is green”) and therefore
with facts about the decomposition of (*). This problem can be solved.
Let us begin with a point that we discussed a moment ago, when we were evaluating
Lewis’ analysis. There is some property – we will refer to that property simply as “#1” –
that a world has in virtue of the fact that (*) is true in it. That very property is had by a
world in virtue of the fact that, in it, any one of the following is true in that world:
(i) D is green and 1+1=2,
(ii) D is green and triangles have three sides,
(iii) D is green and there are infinitely many primes,
and so on.
#1 cannot be identical with (*). This is an immediate consequence of the consid-
erations that we put forth in defense of our rejection of Lewis’ analysis. At the same
time, (*) is true in W iff #1 is instantiated.
Given this, I propose that DP is identical with a set S one of whose members is #1,
and I propose that each of (i)-(iii) is a set that has #1 as a member. I further propose
that for any given one of these propositions to be true is for the members of the cor-
responding set to be instantiated. But, to account for the decompositional differences
holding between any two of the four propositions in question, I will also propose that
no two of these sets have exactly the same members (even though, as we will see, they
share #1 along with some other properties).
For a moment, let us speak as semanticists, and not as metaphysicians. As seman-
ticists, our inclination is to say that “D is green” decomposes into (inter alia) “D” and

8. In this context, we will, for brevity’s sake, ignore the distinctions between properties and
hyper-properties and hyper-hyper-properties…and will simply use the term “property.” Also, in
this context, we will speak rather naively about the nature of property-instantiation and of pro-
pertyhood: we will not always make the necessary allowances for the points developed in the
previous chapter as to the nature of properties. But we will see that this is purely an expository
aid – a way of keeping our discussion from being prohibitively long and tedious, and that there
is no difficulty reincorporating our contentions about properties into our analysis.
 Conceptual Atomism and the Computational Theory of Mind

“green.” Our inclination is thus to say that “D” and “green” are well-formed parts of “D
is green” and, therefore, that these expressions correspond to isolable constituents of
(*), i.e. of what is meant by “D is green.” And, as semanticists, our inclination is to say
that each of these morphemes corresponds to some discrete and ultimate constituent
of what is meant by (*).
We’ve already discussed why the things that we refer to as “individuals” – vases,
plants, people, animals, and so on – can be thought of as properties. That is how we will
think of them in this chapter. Given this, let #2 be the property corresponding to “D”,
and let #3 be the property corresponding to “green.” Finally, let S be a set that contains
all and only #1, #2, and #3.
Given that S comprises #1, it immediately follows that (*) is true if the members of
S are instantiated. In fact, the converse follows as well. If #1 is instantiated, then so are
#2 and #3. So (*) is true exactly if all of #1, #2, and #3 are instantiated.
Notice that S has one part corresponding to “D”, another part corresponding to
“green”, and a third part corresponding to a certain way of “combining” those two parts.
A world instantiates #1 iff it instantiates #2 and #3 and also a third property P that satis-
fies the following two conditions. First, the instantiating of P involves, i.e. is sufficient for,
the instantiating of #2 and #3. Second, the instantiating of both #2 and #3 is necessary,
but not sufficient for the instantiating of P. (Even if D exists in a world W, and the prop-
erty of being green is instantiated in W, it doesn’t follow that D is green in W. It could be
that, in W, D is purple and it is some other object that is green. So, as we said, the instan-
tiating of both #2 and #3 is necessary, but not sufficient, for the instantiating of P.)
(*) is thus a structure that has the following properties. It has an isolable part cor-
responding to D and an isolable part corresponding to the property of being green;
and it also has a part that in some way combines those two parts – that combines them
in such a way that, for (*) to be true, it is not enough that something or other be green
or that D have some property or other, it being necessary that D specifically have the
property of being green specifically.
S has an analogous structure. It has one part corresponding to D, another part cor-
responding to the property of being green, and a third part (#1) that combines the first
two in a certain way. When all of the members of S are instantiated in a given world W,
it follows that three things are the case in that world:
(a) D exists;
(b) there is an instance of the property of being green; and
(c) D itself is such an instance.
Thus, we may naturally identify (*) with S, and (*)’s being true in a given world with
the members’ of S being instantiated in that world.
 Chapter 23.  Conceptual content and the structure of the proposition 

The unity of the proposition

It is obvious that (*) involves some way of “combining” or “synthesizing” D with the
property of being green. The problem is that the nature of this synthesis has never been
identified. (It has always been described to us in empty metaphorical terms, such as
“saturation” and “juxtaposition.”) Given our analysis, we are in a position to identify
the nature of this synthesis.
First of all, when it is said that (*) “consists of ” D and greenness, or (equivalently)
that D and greenness are “combined” in a certain way in (*), the terms “consist of ” and
“combine” obviously don’t denote spatial relations. They instead denote abstract rela-
tions of dependence. Here it may help if we switch to an example that has more rela-
tional structure than (*). (We will come to back (*) shortly.) Consider the proposition:
(**) Smith hit Jones
Obviously (**)’s existence presupposes, and thus depends on, that of Smith and Jones and
the relation of hitting. Further, if (**) is true, then so are the propositions something hit
Jones, and Smith hit something, and also something hit something. The truth of (**) thus
depends in a certain way on facts about Jones, Smith, and the relation of hitting.
There is more to say in this vein. (Here we must keep it firmly in mind that Smith
and Jones, and (so-called) individuals generally, can be thought of as properties.) (**)’s
existence presupposes, not just that of Smith, Jones, and the relation of hitting, but also
that of a certain “synthesis” (whose nature has yet to be described) of these three things.
And (**)’s being true in W involves more than there being instances in W of Smith,
Jones, hitting, Smith’s hitting somebody, and somebody’s hitting Jones. Those instances
must be combined in W. After all, if Smith hit Brown, and Aaron hit Jones, then all of the
properties mentioned will be instantiated in W, but (**) won’t necessarily be true there.
X is a constituent of proposition Y iff Y depends on X in a certain way, in other
words, iff there is some property P such that X’s having P is necessary for Y’s being
true. Of course, as it stands, this statement is much too permissive. For example, the
number two has the property of being a thing x such that snow is white. Therefore,
there is some property P such that 2’s having P is necessary for snow’s being white. But
that number is not a constituent of the proposition that snow is white. Nonetheless,
even though, as it stands, the statement X is a constituent of Y iff, for some P, X’s having
P is necessary for Y’s truth is both wrong and vague, it is still a good jumping-off point
for our analysis of propositional structure. And we will soon replace that wrong and
vague statement with a correct and precise one.
Before we continue, we should deal with an apparent problem with our analysis.
The proposition:
(***) if Smith hit Jones, then Smith hit somebody
can be true without Smith’s hitting anyone or Jones’ being hit by anyone or, in fact, any-
one’s hitting anyone. But notice that this is a molecular proposition, and that, as such, it
 Conceptual Atomism and the Computational Theory of Mind

affirms a relationship not between Smith and Jones, but between propositions. And no-
tice that those two propositions – as opposed to Smith, Jones, and the relation of hitting
– are the immediate constituents of (***). Notice also that, just as the truth of Smith hit
Jones depends on facts about its immediate constituents, so the truth of (***) depends
on facts about its immediate constituents. So given only that (***) doesn’t imply any
particular fact about Smith or Jones or the relation of hitting, there is no reason to reject
our view that constituency-relations are to be understood in terms of relations of de-
pendence. In any case, we will deal with molecular propositions in a moment.
Let us once again look at (*). (*)’s existence presupposes that of D. (Remember
how (**)’s existence presupposes that of Smith and Jones). (*)’s being true in W in-
volves there begin instances in W of D and of the property of being green. (Remember
how (**)’s being true in W involves there being instances of Smith’s hitting somebody,
somebody’s being hit, and so on.)
Now let us look at S in light of these facts. S’s existence presupposes that of D (and,
of course, that of the property of being green). In order for all of S’s members to be
instantiated in W, it is necessary that both D and the property of being green be instan-
tiated there. But, as we discussed, this is not sufficient.
So far, then, there are significant parallels between (*) and S, and also between (*)’s
being true and S’s being such that all its members are instantiated.
After this point, it becomes hard to determine to what extent facts about S parallel
facts about the structure of (*). This is because we know so little about the latter. So far
as we have such knowledge, it is given by vague statements such as those already dis-
cussed, e.g.
(#) where propositions are concerned, the concept of constituency is to be under-
stood in terms of that of the concept of (logico-metaphysical) dependence
But by identifying (*) with S, and (*)’s being true with S’s members all being instanti-
ated, we can fill in these gaps in our understanding, without violating our pretheoretic
intuitions, vague though they may be, as to the nature of propositional structure.
(#) was our starting point. But (#) is obviously unacceptable as it stands. The truth
of (*) is probably constitutively, and not just causally, dependent on facts about physi-
cal law – on facts about protons, electricity, gravitation, and so on. It is very hard to
believe that D – that very plant – could exist in a world governed by different physical
laws or forces, or in a world that didn’t comprise electrons or protons or any of the
micro-particles found in our world. So it is very hard to believe that D could exist in a
world where, for example, there was no gravitation. But clearly the concept gravitation
is not a constituent of (#), at least not in the same sense as the concept green or the
concept D; and the same is true of the concepts electricity, proton, photosynthesis, and
so on. So even though it may be true, even truistic, (#) is not enough; it doesn’t tell us
why D is a constituent of (*), whereas the concept of electricity is not.
But our analysis can help solve this problem. D is a member of S. (When discussing
S’s membership, we referred to D as #2; but obviously that isn’t important.) Similarly
 Chapter 23.  Conceptual content and the structure of the proposition 

the property of being green (#3) is a member of S. By contrast, the concepts of electric-
ity, gravitation, and so on, are not members of S. So by identifying S with (*), we replace
(#) with a precise and clear statement – a statement that agrees with (#), so far as the
latter goes but that, unlike (#), isn’t unclear or otherwise exceptionable.
If S’s members are instantiated, that entails (inter alia) that D is instantiated and
that the property of being green is instantiated. By contrast, it doesn’t entail any state-
ments regarding gravitation, electricity, or photosynthesis – even though, quite possi-
bly, the members’ of S being instantiated is constitutively dependent on facts about
gravitation, electricity, and so on.
In conclusion, if we identify propositions with sets of properties, and a proposi-
tion’s being true with all of its members being instantiated, then we can give clear,
precise, and extensionally correct answers to important questions relating to proposi-
tional structure.

Another illustration of our analysis

Before we deal with molecular propositions, it may be appropriate to give one more
illustration of our analysis of atomic propositions. This is because (*), the proposition
in terms of which we illustrated our analysis, has minimal relational structure; and it
is thus worthwhile to show that our analysis doesn’t break down when applied to rela-
tionally richer propositions. Given this, let us consider (**) once more.
It is pretty clear that Smith, Jones, and the relation of hitting are constituents of it.
It also seems clear (though this point has a more theoretical quality) that (**) also has
as constituents the “propositional functions”: Smith hit y, x hit Jones, and x hit y.
It is also clear that these things must be “combined” in some special way if (**) is
to result: not just any arrangement of those entities will do.
Frege explained why this last requirement must be met. We briefly discussed
Frege’s argument in the last chapter. Here we must develop what we said.
Consider the following list of expressions:
(**F) Smith, Jones, the relation of hitting, Smith’s hitting something y [or: the propo-
sitional function Smith hit y], x’s hitting Jones, something’s hitting something.
(**F) isn’t a sentence; it doesn’t encode a proposition. Further, a sentence will not result
no matter how many referring terms we add to (**F). Suppose that “COM” denotes the
special (and, as of yet, unidentified) way that Smith, Jones, etc. must be combined if
(**) is to result. If we add “COM” to (**F), what results is:
(**F2) Smith, Jones, the relation of hitting, Smith’s hitting something y [or: the prop-
ositional function Smith hit y], x’s hitting Jones, something’s hitting some-
thing, COM.
 Conceptual Atomism and the Computational Theory of Mind

(**F2) is no more a sentence, and no more encodes a proposition, than (**F). Obviously
no expression would result from (**F2) no matter how many more denoting expressions
we added to it. In general, a sentence is not a (mere) list or heap of referring-terms.
What does “Smith hit Jones” have that each of (**F) and (**F2) lacks? “Smith hit
Jones” is grammatical, whereas the other two are not. In “Smith hit Jones”, the expres-
sions are appropriately inflected and they occur in the right order. This is not the case
with respect to either (**F) or (**F2).9
We’ve seen that we can’t turn (**F) into a sentence by adding more referring terms.
We’ve also seen that it is facts about grammar – inflections, word-order, and the like
– that are needed to turn (**F) into a sentence. It seems, then, that grammatical mor-
phemes do not, at least not generally, refer. The ordinal information that they encode
is not to be understood in terms of their referring to anything.
Let us turn our attention back to (**). We’ve seen that among its constituents are:
Smith, Jones, the relation of hitting, Smith’s hitting something [the propositional func-
tion Smith hit y], x’s hitting Jones [the function: x hit Jones], and something’s hitting
something [the function: x hit y]. But what we haven’t seen is how these things must be
combined if a proposition is to result. And – what would seem to be a related point –
we haven’t seen exactly what it is that the inflections in “Smith hit Jones” do. (In this
context, I will use the term “inflection” to refer to all grammatically significant facts –
e.g. word-order, intonation, and so on.10) We’ve seen that they “combine” the constitu-
ents just mentioned in some way or other; we’ve seen that they don’t do so by referring
to anything; and we’ve seen that, as a result of their performing this (as of yet under-
described) feat of “combining”, a proposition is meant by “Smith hit Jones.” But we
haven’t yet put our finger on the identity of this mode of combination. But, in light of
what we said in connection with (*), we may, I think, be able to do so.
Let us begin with Lewis’ insight. There is some property – we will call it simply
“#1*” – that a world has in virtue of the fact that (**) is true in it. For analogues of

9. Here it would be more convenient if we were writing in a richly inflected language, such as
Russian or Latin. It would then be clearer how grammar unifies expressions into more complex
expressions. In English, there is quite as much grammar as there is in Russian. But, to a large
extent, English-grammar is either realized through word-order or is without any acoustic (or
orthographic) representation. So, where English sentences are concerned, its presence is less li-
kely to be discerned than it would be if we were dealing with Russian sentences. But we mustn’t
on that account think that English is less grammar-rich than, say, Russian.
10. I mention intonation because where some languages are concerned, e.g. Mandarin, given
two utterances that differ in intonation but are otherwise identical, one may be grammatical
while the other is ungrammatical. Even in English, it is, at least arguably, ungrammatical not to
end questions with a certain kind of intonation. So in some cases, intonation is a way of rende-
ring grammatical an otherwise ungrammatical utterance. This is not always the case, of course.
 Chapter 23.  Conceptual content and the structure of the proposition 

reasons already discussed, a world W instantiates #1* in virtue of its being the case
that, in W, any one of the following is true:
(i) Smith hit Jones and 1+1=2
(ii) Smith hit Jones and triangles have three sides.
(iii) Smith hit Jones and there are infinitely many primes.
Obviously this list goes on ad infinitum.
So given that (**) is not identical with any of the propositions on that list., (**) is
cannot be identical with #1*. But it is surely suggestive that (**) is true exactly if #1* is
instantiated.
Given this, let S2 be a set one of whose members is #1*. As we’ve discussed, Smith
and Jones can be thought of as properties. Let #2* and #3* be those properties. It goes
without saying that the expression “being hit by Smith” (which corresponds to the
function Smith hits y) and “hitting Jones” (x hits Jones) and also “hits” (x hits y) corre-
spond to properties. (Following tradition, we can think of the relation of hitting as a
property of ordered pairs.) Let #4*, #5*, and #6* be those properties. Finally, let S2 be a
set that has for its members all and only #1*-#6*.
For obvious reasons, if #1* is instantiated, then so are #2*-#5*. And, as we’ve al-
ready discussed, (**) is true (in a given world) exactly if #1* is instantiated (in that
world). It follows that (**) is true in a world exactly if all of #1*-#6* there. So, thus far,
there is no reason not to identify S2 with (**), and no reason not to identify (**)’s being
true with S2’s being such that all its members are instantiated.
Further, S2’s constituency is comparable, in an obvious way, to the constituency of
(**), at least so far as we know anything about the latter. We know that Smith, Jones,
the relation of hitting are “constituents”, in some sense of the word, of (**). We also
have reasons, given to us by Frege (and, since his time, by many others) to suspect that
the functions x hits Jones, Smith hits y are also constituents (in some sense of the word)
of (**). Further, we know that (**) is what results when these (or some subset of these)
various entities are “combined” in some mysterious way. S2 has a constituent corre-
sponding to each of the constituents of (**), and the mysterious operation of “combin-
ing” just mentioned can be understood in terms of facts about S2’s membership.
S2 has a discrete part corresponding to each of Smith, Jones, the relation of hitting,
and so on. And we know exactly how #1*-#6* – the parts corresponding to the Smith,
Jones, and so on – relate to S2: they are members of it. That is as clear a relationship as
we can hope to find. So, thus far, facts about S2’s composition correspond to what we
know about (**)’s composition – with the qualification that, in the case of S2, it is ex-
tremely clear what those compositional facts are.
We know that for Smith, Jones, and so on, to be “constituents” of (**) is for the
truth of the latter to have some kind of dependence on facts about the former. But we
don’t know what exactly that kind of dependence is. This problem vanishes if we iden-
tify (**)’s being true with S2’s being such that all its members are instantiated. If all S2’s
 Conceptual Atomism and the Computational Theory of Mind

members are instantiated in W, then Smith hits somebody in W, Jones is hit by some-
body in W – and, of course, somebody hits somebody in W, not to mention that Smith
and Jones exist in W. Similarly, (**)’s being true in W involves Smith’s hitting some-
body in W, Jones’ being hit by somebody in W, and so on. Our analysis is therefore
consistent with everything we know about the dependence-relations that hold between
propositions and their constituents.
At the same time, as we discussed, not all dependence-relations correspond to
relations of propositional constituency. Smith’s hitting Jones is (we may suppose) con-
stitutively dependent on various facts about neurons. But the concept neuron isn’t a
constituent of (**). Further, if Smith hit Jones, it follows logically (though not formally)
that Smith put some kind of pressure on Jones’ body. But the concept pressure isn’t a
constituent of (**). So it isn’t enough to say: X is a constituent of proposition Y iff Y’s
being true involves, or is otherwise metaphysical dependent on, some fact about X.
Though true, that statement is too broad, and it doesn’t say why the concept of a neu-
ron is not, whereas the property of hitting is, a constituent of (**).
But this problem vanishes if we identify (**) with S2. The concepts (or properties)
neuron and pressure are not members of S2. So if we identify (**) with S2, it is clear why
the concepts neuron and pressure are not constituent of (**), even though states of af-
fairs instantiating those properties are metaphysically or logically necessary for the
truth of (**). Further, if all of S2’s members are instantiated, that entails that somebody
or other hit Jones, but it doesn’t entail anything having to do with neurons.

An apparent problem for our analysis

Here it might seem that we run into a problem. If S2’s members are all instantiated, that
does entail, at least arguably, that Smith put some kind of pressure on Jones and that
Smith thus altered the balance of forces acting on Jones. But the concepts pressure, bal-
ance of forces, and so on, are not constituents of (**). So it might seem that there has
been a breakdown in our attempt to understand (**)’s structure in terms of facts about
what is entailed by the supposition that all of S2’s members are instantiated.
But there hasn’t really been a breakdown. There is an obvious sense in which the
truth of Smith hit something is, whereas any proposition about pressure or force is not,
an immediate consequence of the supposition that S2’s members are instantiated. One’s
making the latter supposition is identical with (inter alia) one’s supposing that Smith
has the property of hitting something, that Jones has the property of being hit by some-
thing, and so on. By contrast, one’s supposing that S2’s members is obviously not identi-
cal with one’s making suppositions about alterations of the balance of forces involving
Jones. So even though the supposition that S2 is instantiated entails both Smith hit
something and also Smith altered the balance of forces acting on something, these two
entailments differ from each other in some fundamental respect. Our analysis enables
us to say exactly what that difference is. Indeed it practically amounts to a description
 Chapter 23.  Conceptual content and the structure of the proposition 

of that difference. Our analysis therewith enables us to say why the property of being a
force is not, whereas the property of hitting is, a constituent of (**).

The bearing of our analysis on SCT

Before we discuss molecular propositions, let us discuss the bearing that our analysis of
propositionality has on the viability of SCT. If our analysis is right, a proposition is a set
of properties. Given what we said earlier, this means that any proposition is at an extraor-
dinarily high number of removes from anything that could be the content of any experi-
ence (i.e. any perceptual or sensory experience). Let us now spell out why this is so.
We’ve already seen why, so far as they cannot be identified with concrete situations,
property instances are abstract commonalities holding among situations, and are there-
fore abstract entities (contrary to what is generally thought). We’ve also discussed why
terms like “red” are not properties of things whose instances can be encountered in ex-
perience, but are rather properties of such properties. We had to invent special terms –
“R1”, “R2”, and so on – to denote the maximally specific shades of red that one sees. What
is true of “red” is true of any expression with a fixed, as opposed to context-dependent,
reference. Context-dependent expressions, e.g. “that exact shade of red”, can denote
properties had by situations that can be experienced, i.e. seen, heard, touched, and so on.
But context-dependent expressions must be built of context-independent ones; and the
latter, as we’ve discussed, necessarily have abstract entities for their significances.
We’ve also discussed why terms like “Smith” and “Jones” are not as dissimilar as
one might think from terms like “red” and “pain.” The former, no less than the latter,
fail to have concrete situations for their semantic contents, and instead have for their
semantic contents abstract commonalities holding among situations or worlds.
If our analysis of propositions is correct, then a proposition is a set consisting of
various properties. And in most cases those properties are really hyper-properties
(properties of properties, or properties of properties of properties, or…) The only cases
where the constituents of a proposition are mere properties, as opposed to hyper-prop-
erties, are the cases where those propositions are to be expressed using demonstrative
expressions like “that particular shade of red” (or neologisms like “R1”). For a proposi-
tion to be true is for the properties that are members of it to be instantiated, i.e. it is for
all of those (hyper-)properties to have the higher-order property of being instantiated.
 Conceptual Atomism and the Computational Theory of Mind

Molecular propositions

Earlier we saw that molecular propositions might pose a problem for our analysis. Let
us allow an imaginary objector to articulate exactly what that problem would be:
Consider the proposition:

(N**) “Smith did not hit Jones.”

Surely (N**) has Smith, Jones, and the relation of hitting as constituents. In fact,
(N**) has all of the constituents had by (**). (At the same time, (N**) has an extra
constituent, corresponding to the operation of negation.) But for (N**) to be true
is not for those properties to be instantiated. Indeed, it is for those properties, or
at least some of them, not to be instantiated. This makes it questionable whether
you have analyzed propositions correctly and, therefore, whether there is any
merit to your anti-SCT use of the concept of a proposition.

(N**) is a molecular proposition. It is the proposition: it is not the case that Smith hit
Jones. (N**) has two immediate constituents. These are (**) and the concept of nega-
tion. Jones is not an immediate constituent of (N**). (N**) is a statement about (**).
(N**) says of some proposition that it isn’t true. If our analysis is correct, this is equiv-
alent to saying that the members of S2 are not instantiated.
At the same time, we want to hold onto the idea that a proposition is a set of prop-
erties, and that for a proposition to be true is for the members of that set to be instanti-
ated. In other words, we don’t want to understand the concepts proposition and truth
one way where atomic propositions are concerned, and a different way where molecu-
lar propositions are concerned. So we must show that (N**) is some set SN of proper-
ties, and that (**N) is true iff all of SN’s members are instantiated. This can be done.
Remember #1* – the property had in common by all and only those worlds where
(**) is true. As we discussed, #1* is a certain way that a “quantum” can be rippled. Now
consider all the worlds that aren’t rippled that way – that have ripples incompatible
with the truth of (**). Let #N1 be that property.
Of course, in this context we described #N1 negatively. We described it in terms of
some proposition’s not being true. But, to echo a point made by Frege (1918), it doesn’t
follow that #N1 is itself any more “negative” than #1. Any given state of affairs can be
described positively or negatively. We can say “Brown failed to make it to the finish
line” (negative characterization) or “Brown collapsed before the finish line” (positive
characterization). The first sentence describes a state of affairs in terms of the falsity of
some proposition (Brown made it to the finish line); the second describes that same
state of affairs but not in terms of the falsity of some proposition. Similarly, suppose we
were so naturally aggressive that the normal course of events was to hit other people,
and that it was more disruptive to our psychologies and cultures when people refrained
from hitting others than when they actually did so. In that case, we might have a spe-
cial verb “to frit”, meaning to refrain from hitting. So “Smith frit Jones” would mean:
 Chapter 23.  Conceptual content and the structure of the proposition 

Smith refrained from hitting Jones (despite there being innumerable internal and exter-
nal pressures acting on Smith that would tend towards his hitting Jones). In that case, the
state of affairs that, in actuality, is expressed by (**) would be expressed negatively, i.e.
by the sentence “Smith did not frit Jones.” So #N1 is quite as “positive” a property as #1,
and there is no threat of vicious circularity in our use of “Smith hit Jones” to identify
#1. This is because we were using that sentence only to identify that property; and, as
we just saw, that same property could have been identified without using the sentence
“Smith hit Jones” or any synonym. Given this point, let us move on.
Let SP be the property of being identical with S2 or, if you prefer, with (**). The prop-
erty of being identical with a set can be instantiated even if the members of that set are not.
The property of being identical with is instantiated in any world where that set exists, it
being irrelevant whether the members of that set are instantiated. Let Z be the set whose
sole member is the property of being a gold object in New Zealand weighing more than a
million pounds. Z exists in this world (and, arguably, in every world). The property of
being identical with Z is instantiated in this world, even though its member is not.
Finally, consider the property of not being instantiated. This is, of course, a prop-
erty had by properties. (It is had by the property of being a mansion on Earth made out
of solid gold.) It is thus a higher-order property (in fact, it is of an even higher-order
than most of the properties we have considered). Let NEG be this property.
Let S3 be the set that consists of NEG, Z, and N1. (N**) is true in all and only those
worlds where N1 is instantiated. Given any world where N1 is instantiated, it follows,
for obvious reasons, that each of Z and NEG is instantiated. So (N**) is true in a given
world iff all of S3’s members are instantiated in that world. Further, S3’s membership
corresponds with the decomposition of (N**), at least in so far as the latter is perspicu-
ously represented by sentences expressing that proposition, e.g.
(&) “it is not the case that Smith hit Jones.”
(&) has two immediate, proper constituents, namely: “that Smith hit Jones” and “it is not
the case.” So if we count “it is not the case that Smith hit Jones” as a constituent (an im-
proper one) of itself, (&) has three immediate constituents. Similarly, S3 has three mem-
bers; and each of these members corresponds in an obvious way to the three main con-
stituents of (&). So facts about S3’s membership neatly correspond to facts about the
decomposition of (&) and therefore, presumably, to that of (N**); and facts about the
conditions under which S3’s members are instantiated correspond neatly to facts about
the conditions under which (N**) is true. So there is every theoretical incentive to iden-
tify (N**)’s being true with its being the case that the members of S3 are instantiated. Thus,
far from warranting the rejection of our analysis, consideration of (N**) confirms it.
What we said about (N**) is true mutatis mutandis of all molecular propositions.
I discuss this at length in another work (Kuczynski 2005c), where I also discuss the
structure of analytic propositions. But given that our concern here is mental represen-
tation – specifically, the viability of SCT – it isn’t necessary that we consider the con-
cept of propositional structure in such detail. And given only what we’ve said thus far
about propositions, the implausibility of SCT has been adequately demonstrated.
 chapter 24

Peacocke on concept-possession

Like us, Christopher Peacocke rejects conceptual atomism. Further, Peacocke has ably
defended a version of non-atomism. The purpose of this chapter is to evaluate Pea-
cocke’s analysis. To do this, we will have to use some of the points that we’ve developed
over the last two chapters. That is why we were not able to discuss Peacocke’s important
work in an earlier chapter.
Let us start with a point about terminology. In Chapter 16, we distinguished be-
tween “conceptso” and “concepts.” The latter are psychological entities; the former are
platonic abstracta. In this chapter, we will once again use these terms in this way.
Let there be some concepto C satisfying the following conditions. A belief in P
coupled with a belief in Q leads directly, i.e. not by way of an inference, to a belief in P
C Q; and a belief in P C Q leads directly to a belief in P and a belief in Q. In other words,
given a belief in P and in Q, one is “primitively compelled”, as Peacocke puts it, to be-
lieve P C Q; and given a belief in P C Q, one is primitively compelled to believe P and
also to believe Q.
Under these circumstances, says Peacocke, we may conclude that C is the con-
cepto of conjunction and also that you grasp that concepto. The concepto of conjunction
has certain “possession conditions”, as Peacocke puts it, and you grasp that concepto
because you satisfy those conditions.1
It is obvious how to apply Peacocke’s analysis to various other conceptso. Let C be
a concepto satisfying the following conditions. If you believe that x falls under C, then
you are primitively compelled to believe x is a closed figure, x has three-sides and x’s
sides are straight; and if you believe all of those three propositions, then you are primi-
tively compelled to believe x falls under C. In that case, says Peacocke, C is the con-
cepto of a triangle.
There are some apparently compelling reasons to accept this analysis. First of all,
intuition recoils at the notion of someone who believes that x is a triangle but has no
disposition to believe that x is three-sided. To believe that x is a triangle just is (so it
would seem) to believe (inter alia) that x is three-sided. This suggests that one’s grasp
of a concepto necessarily involves a mental representation of the fact that certain tran-
sitions are valid. This representation cannot be identical with a tendency to infer that x

1. Here is another illustration of Peacocke’s analysis. Let there be some concepto C satisfying
the following condition. For any proposition P, if you believe P, then you are primitively compel-
led to disbelieve CP (read: “P falls under C”). Under those circumstances, C is the concepto of
negation, and you grasp that concepto because you satisfy its possession-conditions.
 Conceptual Atomism and the Computational Theory of Mind

is three-sided (or closed or planar…) from the fact that x is a triangle, since the ability
to make such an inference presupposes a grasp of the concepto of triangularity. For the
same reason, this mental representation cannot consist in one’s knowledge that x is a
triangle entails x is closed (or x is planar…) Given this, it seems reasonable, if not de
rigueur, to suppose that this mental representation is constituted by a set of disposi-
tions of the kind described by Peacocke – inferentially unmediated dispositions to
transition to certain beliefs, given certain other beliefs.
Fodor (1998: 74) makes the following objection to this line of thought (this is a
paraphrase, not a quotation):
You say that “intuition recoils” at the view that one could believe that x is a triangle
without believing, or at least being disposed to believe, that x is three-sided. But that
intuition may just reflect the fact that it is obvious, at least to us, that triangles have
three sides. If you know that your friend Smith is miserable whenever it rains, then
you are likely to believe that at least one person is miserable if you already believe
that it is raining. But it would be absurd to say that your concept of Smith was con-
stituted by this disposition. And given only that you are likely to believe that x is
three-sided if you already believe that x is a triangle, it is comparably absurd to sup-
pose that your concept of triangularity consists in your having this disposition.

But in light of what Peacocke says about conceptso, there is a compelling response to
this. If x is identical with Smith, that is not in virtue of the fact that x is unhappy when
it rains. But if C is the concepto of triangularity , that is in virtue of the fact that a thing
x falls under C only if x also falls under the conceptso three-sided figure and figure
whose interior angles add up to 180°. . So, whereas Smith is not individuated by the fact
that he is unhappy when it rains, the concepto of triangularity is individuated by the
fact that it has these analytic liaisons with these other conceptso. So for C to be the
concepto of triangularity just is for C to be such that anything falling under it falls un-
der these other conceptso. Given this, it makes little sense to suppose that one could
grasp the concepto of triangularity without mentally representing at least some of these
entailments. For the reasons given earlier, it thus seems necessary to suppose that a
grasp of that concepto consists in a set of “inferentially unmediated” dispositions of the
kind that Peacocke describes.
In a word, if we give any weight to the intuition that a grasp of a concepto is in-
separable from an appreciation of the validity of certain patterns of inference, it cer-
tainly appears that we have little choice but to accept Peacocke’s analysis.
George Rey put forth some criticisms of Peacocke’s view. I will discuss what I believe
to be Rey’s two most significant criticisms. One of those criticisms is spurious. But the
other is cogent, and Peacocke’s analysis must therefore be rejected. We will also find rea-
sons of our own to reject Peacocke’s analysis, as well as the arguments just provided for it.
 Chapter 24.  Peacocke on concept-possession 

Rey’s first criticism (the spurious one)

Since Peacocke’s analysis sees conceptso as being individuated by their analytic struc-
tures, it obviously presupposes that there is an analytic-synthetic distinction. Quine
famously denied that there is an analytic-synthetic distinction, and many agree with
Quine on this. So, as Rey points out, Peacocke’s position is wrong if Quine is right. Rey
believes that Quine is right, or at least that he may be right, and he therefore holds that
Peacocke’s position is either false or tenuous.
As we saw in Chapter 16, Quine is not right. Nonetheless, the basic idea behind
Rey’s criticism is not an unreasonable one. But the inadequacies that Rey ascribes to
Peacocke’s position are better understood in terms of Kripke’s work than they are in
terms of Quine’s. Peacocke’s position demands not only that some conceptso have some
analytic structure, but that any two non-identical conceptso have different analytic struc-
tures. So the concepto Socrates must have a distinct structure from the concepto Plato. We
know from Kripke that, with a few irrelevant qualifications, “Socrates was bald” doesn’t
entail anything not entailed by “Plato was bald.” (Obviously “Socrates was bald” entails
“Socrates was bald”, whereas “Plato was bald” does not; and the one sentence does,
whereas the other does not, entail “either Socrates was bald or there are square circles.”
But for reasons discussed earlier these differences fail to show that the proposition meant
by “Socrates was bald” doesn’t entail anything not entailed by the proposition meant by
“Plato was bald.” Hence these differences in entailment-relations are irrelevant in this
context.) Given this, a case can certainly be made that the conceptso corresponding to
“Socrates” and “Plato” do not differ in respect of analytic structure.
In connection with this, it should be pointed out that Peacocke always illustrates
his position in terms of conceptso belonging to arithmetic or mathematical logic (e.g.
triangle and disjunction) he never relates his analysis to conceptso like bird or Socrates
or tree. Given this, it is hard to escape the feeling that Peacocke’s position fails to hold
for anything other than a few artifacts and that it therefore doesn’t hold at all.

Evaluating this criticism

This criticism presupposes the truth of a conception of conception and of linguistic un-
derstanding that we have found to be indefensible. Kripke is right that (with the irrele-
vant exceptions identified earlier) FSocrates has phiG doesn’t entail anything not also en-
tailed by FPlato has phi.G But, as we saw in Part I, we cannot read facts about conception
and mental representation off of facts about literal meaning. Having a concept of Socra-
tes consists precisely in knowing the truth of some description that singles out Socrates.
I grasp the literal meaning of “Socrates was wise” through an existence-claim; I grasp
“Plato was wise” through another existence-claim; and those claims are very different in
terms of their analytic structures. When we look at what is actually grasped, we find that
 Conceptual Atomism and the Computational Theory of Mind

we don’t have an example of two distinct propositions that have the same analytic struc-
ture. We find two different propositions that have different analytic structures; and we
find that, so far as those propositions differ, it is in virtue of the fact that one of them is
about Socrates while the other is about Plato, i.e. it is in virtue of the fact that, in the place
where one of them comprises a concepto that Plato uniquely instantiates, the other com-
prises a concepto that Socrates uniquely instantiates. So, as we just said, we have two
different conceptso that have very different analytic structures. This is exactly what Pea-
cocke’s analysis needs. So Rey hasn’t yet exposed any real problems with that analysis.
Also, as we discussed in Chapter 11, the very idea of two propositions having the
same analytic structure is of questionable coherence. (Let us use the word “content” to
denote either propositional or non-propositional information.) Suppose that contents
c1 and c2 are such that, for any content c3, you can validly infer c3 from c1 in manner M
(i.e. by way of such and such rules of inference, applied in such and such order) exactly
if you can validly infer c3 from c2 in manner M. In that case, c1 and c2 are one and the
same. Rey’s criticism implicitly involves a rejection of this presumption and is therefore
false. So we may conclude that Rey’s first criticism of Peacocke’s view has little force.

Rey’s second criticism (the cogent one)

Let us start with a story. Smith believes that n=32. He non-inferentially transitions
from that belief to a belief that n=9. Trivially, there are two different views to take vis-
à-vis Smith’s transition. On the one hand, we can say that it was either valid or invalid.
On the other hand, we can say that it was neither valid nor invalid.
Let us elucidate this second option. If I think Socrates was wise at 3:00 p.m., and
think something was wise at 3:01 p.m., I haven’t necessarily made a valid inference. In
fact, I haven’t necessarily made a valid inference even if my thinking Socrates was wise
is what caused me to think something was wise. This can be shown in terms of a story
that we told in Chapter 15. A hypnotist programs me to believe something was wise
whenever my heart-rate goes above 100 BPM. When I discover that Socrates was wise,
the resulting epiphany leads to tachycardia and thus, given the hypnotist’s program-
ming, to my forming the belief that something was wise. Here no inference occurred.
By the same token, if I believe that something was wise at 3:00 p.m., and a moment
later form the belief that nothing was wise, I am not necessarily guilty of inferring a
proposition from its negation. This can be shown through an obvious adaptation of the
story just told.
For similar reasons, given only that Smith transitions from a belief that n=32 to a
belief that n=9, it doesn’t follow that his transition is valid; and if he transitions from a
belief that n=32 to a belief that n=6, it doesn’t follow that his transition is invalid (even
though it does follow that he was wrong, at least once, as to n’s identity).
If we say that Smith’s transition is either valid or invalid, then we are saying that it
is answerable to norms. More specifically, we are saying that it must be consistent with
 Chapter 24.  Peacocke on concept-possession 

the content of Smith’s belief that n=32 and also with the content of various general be-
liefs on Smith’s part (e.g. that no one number can be identical with two distinct num-
bers). But in that case, Smith’s making that transition cannot be constitutive of his be-
lieving that n=32. In general, since there is normativity only where there is already
fixed content, norm-governed transitions cannot be constitutive of content.
So if Peacocke’s analysis is to be correct, the transitions of which he speaks must
not be answerable to any norms. The problem is that at least some transitions that are
not norm-governed are not constitutive of anyone’s grasp of any concepto. The story we
told a moment ago shows this. At 3:00 p.m. I form the belief that Socrates was wise,
and I non-inferentially transition to a belief that something was wise. Given only this
much information, it doesn’t necessarily follow that an understanding of existential
generalization is implicated in that transition. It could be that I am only carrying out
the hypnotist’s instructions. So if Peacocke’s position is to hold up, he must say that
some, but not all, non-inferential transitions are constitutive of conception.
But how are the right non-inferential transitions to be distinguished from the
wrong ones? Are the mechanisms underlying the right ones sturdier than those under-
lying the wrong ones? Not necessarily. It could be that the mechanism involved in my
carrying out the hypnotist’s instructions is as sturdy as a connection can be. Are the
right transitions those that happen reliably? Not necessarily. It could be that every non-
inferential transition that I make is the result of post-hypnotic suggestion. The differ-
ence between a principled and an unprincipled transition cannot be understood in
purely statistical terms. Speaking statistically, unprincipled transitions may be much
more common than principled transitions.
One might counter-respond by saying that all other things being equal one is more
likely to make a principled than an unprincipled transition. But this is to little avail. In
this context, the term “all other things being equal” seems to mean: “so long as one isn’t
under hypnotic suggestion or any other compulsion which results in an unprincipled
transition.” So the counter-response in question becomes the innocuous claim that
one makes principled transitions – except when one doesn’t.
There thus doesn’t seem to be any non-arbitrary way of distinguishing the non-
inferential dispositions that are constitutive of conception from those that are not thus
constitutive. Peacocke’s theory is therefore either arbitrary or false. It is arbitrary if it
says that only some non-inferential dispositions are constitutive of conception, and
false if it says that all of them are.

Other problems with Peacocke’s view

According to functionalism, brain-state B’s realizing a belief that Socrates was wise
consists in its having certain consequences, e.g. its leading to a belief that something
was wise. But if that is right, then B’s realizing such a belief cannot be what generates
those consequences.
 Conceptual Atomism and the Computational Theory of Mind

Peacocke’s analysis is a form of functionalism. It is in virtue of its having certain

effects that B realizes a belief that Socrates is wise. But in that case, it is hard to see why
the just-mentioned problem with functionalism isn’t a problem for Peacocke’s view.
Admittedly, Peacocke doesn’t hold that all of the states of affairs generated by B
determine what content it has. He seems to hold that the only relevant effects are those
that B immediately generates.
But this doesn’t help. Suppose that B leads to B2, which realizes a belief that some-
thing was wise, and that B2 leads to B3, which realizes a belief that something was
sentient. If Peacocke’s analysis is right, B2’s having that effect is constitutive of its being
a belief that something was wise. So B2’s having that effect is constitutive of something
that is constitutive of B’s being a belief that Socrates was wise. For similar reasons, B3’s
being a belief that something was wise is constitutive of something that is constitutive
of B’s being a belief that Socrates was wise. At any given juncture, we encounter, at
most, something that is constitutive (either directly or indirectly) of B’s being a belief
that Socrates was wise. At no juncture, therefore, do we encounter anything that could
be an effect of B’s being such a belief. Never do we encounter anything that is an effect
of my belief that Socrates was wise. Peacocke’s analysis must be wrong, given that my
believing that Socrates was wise obviously does have effects.
We must now deal with an altogether different problem with Peacocke’s analysis. If
Peacocke is right, conceptso (notice the subscript) are individuated entirely by their ana-
lytic liaisons. So conceptso are relations to other conceptso. Any given concepto is a node
in a network that, in its turn, consists in nothing but more conceptso.
It is hard to see how this picture can allow conceptso to make any contact with any-
thing non-conceptualo. Right now, let us suppose, physicist Smith is having various
perceptions. On the basis of these perceptions, along with a correct understanding of
the conceptso electron, proton, weak force, and so on, Smith makes a correct and war-
ranted inference as to the sub-atomic structure of object X. This shows that, if such and
such macroscopic conceptso are instantiated, that provides good evidence that thus and
such microscopic conceptso are instantiated. And this, in its turn, confirms the general
point that what we perceive is sometimes relevant to how we ought to see conceptso as
being interrelated. We saw in Chapters 22–23 that perceptual content is not cenceptualo
and therefore, that perception is not conception. And we saw a moment ago that per-
ceptual (and therefore non-conceptualo) mentation gives us information as to how con-
ceptso how are interrelated. It follows that not all relations among conceptso are to be
understood in strictly conceptualo terms. But Peacocke’s analysis demands precisely
that conceptso be understood strictly in terms of other conceptso. For, according to that
analysis, a concept O is a node in a structure that consisted of nothing but conceptso.
Hilbert famously gave purely formal definitions of terms like “line”, “point”, “paral-
lel with”, and so on. Each of these terms was defined in terms of the others, and none
was defined in terms of anything that had any sensory or para-sensory content. So
long as one thinks of the totality of conceptso as forming a purely formal structure,
such as Hilbert’s formalization of geometry, one may not be guilty of saying that any
 Chapter 24.  Peacocke on concept-possession 

given concepto is to be understood solely in terms of its relations to other conceptso.

But conceptso are not purely formal entities. There are right and wrong ways to con-
ceptualize experience. I am right to think that plant is green, wrong to think that plant
is red, and (supposing that I am still looking at that same plant) even more wrong to
think that Martian is red. If conceptso were to be understood strictly in terms of their
mutual relations, no conceptualization of any experience would be any better or any
worse than any other.
Conceivably one might say the following in response to this:
Any experience can be conceptualized in any way that one wishes. It is not conceptso
per se that rule out certain conceptualizations of experience. It is conceptso plus cer-
tain auxiliary hypotheses that do so. Those auxiliary hypotheses link conceptso to
experience and thus provide experiential models of them, just as statements about
light-beams bouncing off mirrors can model statements about triangles, i.e. they pro-
vide “physical interpretations” of geometrical truths.

But this point of view rebuts itself. It is a datum that I am right to think grass is green
and wrong to think penguins wear top hats: given the sensory experiences I have had,
and given the structures of the conceptso grass and green and so on, I could not reason-
ably believe that grass is purple. It isn’t merely that such a belief would be unreasonable
relative to arbitrarily chosen auxiliary hypotheses. It is rather that any auxiliary hy-
potheses that, given my sensory experiences, warranted my believing that grass is pur-
ple would ipso facto run afoul of the structures of the conceptso grass and green and so
on. So the auxiliary hypotheses that the objector mentions must themselves be internal
to the structure formed by the totality of all conceptso.
Of course, what one ought to believe on the basis of a given perception is a func-
tion of what one already believes. If I believe that I am looking at myself in a normal
mirror, perception P warrants the conclusion that I am eleven feet tall. If I believe that
I am looking at myself in a mirror that is warped in a certain way, perception P war-
rants the view that I am six feet tall. (So under those circumstances, I must conclude
that I am looking at myself in a warped mirror.) But if any experience, or series of ex-
periences, can be correctly conceptualized in any way at all, then it won’t matter what
other experiences I’ve had. For, under that circumstance, those other experiences will
fail to provide a benchmark in terms of which I am to know how to conceptualize P. So
to the extent that P warrants the view that I am of this as opposed to that height, it is
because experiences other than P are to be conceptualized in certain ways and not in
others. It is therefore not an option to say that the principles linking conceptso to expe-
rience are external to those conceptso themselves.
Whether a concepto is instantiated is typically to be decided on experiential grounds.
(I say “typically” because where some conceptso are concerned, e.g. even number and
square circle, it can be known non-perceptually whether they are instantiated or not.) It
is only on the basis of sense-perception that I can know whether somebody is wearing a
blue-shirt, i.e. it is only on the basis of sense-perception that I know the concepto blue-
 Conceptual Atomism and the Computational Theory of Mind

shirt wearer to be instantiated. But in order for experience to warrant such a belief, the
concepto blue-shirt wearer must already be such that certain kinds of experiences warrant
the view that it is instantiated. (Of course, in this context, “warrant” means “inductively
warrant.”) Even though it is an empirical matter whether a given concepto is instantiated,
it is not ultimately an empirical matter what kinds of experience would warrant the view
that it is instantiated. Given only the structure of the concepto blue-shirt wearer, it doesn’t
follow that somebody is wearing a blue shirt. But given the structure of that concepto, it
does follow that if one were to have certain sense-experiences, one would be right to
believe that concepto to be instantiated.
Most empirical judgments can be overturned by later experiences. (I say “most”
because, arguably, some judgments about phenomenology are not corrigible.) But this
proves only the trivial point that empirical judgments are not analytic. It doesn’t show
that it is not analytic whether a certain kind of experience would confirm a certain
empirical judgment. In fact, it shows the opposite. So far as an empirical judgment is
capable of being overturned by future experience, that is because, given the conceptso
involved in the proposition affirmed by that judgment, it is already determined what
experience would have to be like in order to warrant a rejection of that judgment.
If Peacocke’s analysis is right, then any given concepto’s structure is to be under-
stood strictly in terms of that concepto’s relationships to other conceptso. Peacocke’s
analysis is therefore inconsistent with the fact that there are right and wrong ways to
conceptualize experience; and it is thus inconsistent with the related fact that, ulti-
mately, whether a certain kind of experience would warrant an empirical judgment is
not itself an empirical matter.
To evade this charge, Peacocke would have to say that sense-experience was itself
some kind of conceptualization. But this is not a reasonable position. (Nor would Pea-
cocke take it. Peacocke is acutely aware of the distinction between conceptualo and
non-conceptualo content, having explored that distinction with more subtlety than any
other author, and he rightly holds that perceptual content is non-conceptualo.2) My
thinking that Tom is wearing a blue-shirt is very different from my seeing a state of
affairs that involves Tom’s wearing a blue shirt. The first is a conceptualization of vari-
ous experiences. The second is not.3
Also, if we say that sense-experience is some kind of conceptualization, then we
are left with the question – a conceptualization of what? If experience is to provide any

2. See Peacocke (1989, 1992).

3. Given that the former is itself an experience, it would be viciously regressive to say that it
was itself, in its entirety, a conceptualization of experiences.
Also, If we say that sense-experience is a conceptualization of sorts, then we will only ren-
der the term “conceptualization” ambiguous. If we came up with a new pair of terms to mark the
distinction that we just obliterated, then the very problems that we’ve been discussing would
arise in connection with that pair. This shows that we are simply re-labeling conceptso, not ana-
lyzing them, by saying that my seeing Tom is a case of conceptualization.
 Chapter 24.  Peacocke on concept-possession 

basis for any conceptualizations, then experience cannot itself be conceptual, lest it be
reduced to an epistemically inert “frictionless spinning in a void”, to use Davidson’s
(2001: Chapter 10) expression.
McDowell (1994) argues that, because sense-experience warrants certain concep-
tualizations, it must itself be conceptual. McDowell is right to emphasize that a theory
of experience must allow that facts about conceptualo structure determine what judg-
ments we can legitimately make on the basis of a given experience (or set of experi-
ences). But it doesn’t follow that sense-experience is itself nothing other than a kind of
incessant conceptualization. And we’ve already seen one reason to think it absurd to
suppose that experience is itself conceptual.
Here is another such reason. To conceptualize one’s experiences is to discern ab-
stract commonalities holding among them. It follows that it would be absurd to main-
tain that experience is itself a process of incessant conceptualization. It follows, indeed,
that the content of experience is fundamentally different from the content of a concep-
tualization. Denying this would be like saying that similarities among people are them-
selves people.

An alternative to Peacocke’s analysis

Let us start with some facts. I am disposed to believe x has more than two sides if I
believe x is a triangle. I am disposed to believe m is larger than n if I believe m is what
results when some positive number is added to n. Given a grasp of any concepto C, one’s
beliefs tend to track C’s analytic structure. Peacocke’s analysis is meant to accommo-
date this important fact, and at first it seems to do a reasonable job.
But we’ve also seen that Peacocke’s analysis is false. What should replace it? We
need to produce a theory that, like Peacocke’s, accommodates the datum just men-
tioned but that, unlike Peacocke’s, doesn’t strip the mental causal potency and that,
unlike Peacocke’s, doesn’t strip conceptso of experiential import.
I believe that the elements of a solution may lie in some points that we made in the
last two chapters. First of all, conceptso are not so different from properties. For x to
have the property of triangularity just is for x to fall under the corresponding concepto.
In fact, for many authors, the words “property” and “concept” are interchangeable.
Admittedly, this line of thought must be qualified. The concepto successor of one is
obviously different from the concepto predecessor of three. (This is because thinking x is
a successor of one is different from thinking x is a predecessor of three.) But the prop-
erty of being a successor of one doesn’t seem to be different from the property of being
a predecessor of three. (This is because anything that has the one property ipso facto
has the other: so being a successor of one is not different from being a predecessor of
three.) Nonetheless, it does seem at least approximately true to say that conceptso just
are properties. If only to facilitate exposition, let us proceed on the assumption that
this is right. (Later we will argue that this assumption is correct.)
 Conceptual Atomism and the Computational Theory of Mind

In light of this, let us remember what we said about properties. Property-instances

are abstract objects, contrary to what has generally been assumed. Properties them-
selves are commonalities holding among such abstract objects. The properties that are
expressed by most natural-language predicates – e.g. “red”, “hot”, “square-shaped” – are
commonalities holding among these commonalities: they are hyper-properties. The
only natural-language expressions that denote bona fide first order properties are to-
kens of certain demonstrative expressions (e.g. “the exact color of that rose”).
These metaphysical points have epistemological parallels. Representations of
property-instances are not cognitively fundamental. Representations of situations are
fundamental; and property-instances are understood in terms of situations. Property-
instances are commonalities holding among situations, and representations of prop-
erty-instances are representations of such commonalities.
This suggests that our concepts of properties are descriptive in nature. But, ulti-
mately, our concepts describe properties not in terms of other properties, but in terms
of situations. To grasp a property is to see what situations have in common. (More
precisely, it is to see what is had in common by commonalities that hold among com-
monalities….that hold among situations.) A cognitive representation of a property just
is an understanding of what different situations (or commonalities that hold among
commonalities…that hold among situations) have in common. We grasp properties by
knowing descriptions that they satisfy.
Our analysis is consistent with the fact that conception tracks analytic structure.
But this structure will not always be represented propositionally. This is because grasp-
ing conceptso (or properties) is obviously a prerequisite to grasping propositions, given
that propositions are built out of conceptso and given that one cannot grasp proposi-
tions without grasping their constituents.4
Of course, there are some conceptso such that, to grasp them, one must first grasp
some proposition or other, e.g. the concepto of being a thing x such that either x is tall
and snow is white or x is short and snow is green. But it is viciously regressive to say
that conceptso must always be represented in terms of propositions. Ultimately con-
ceptso are understood in non-propositional terms. For similar reasons, they are (ulti-
mately) understood in non-conceptualo terms.

4. Fodor (1990, 1998) argues that one’s grasp of a conceptO – e.g. one’s grasp of the conceptO of
justice or redness or sentience – is completely non-descriptive. In Fodor’s view, for me to have a
concept of justice just is for some instance of justice to have a certain causal connection to my
present neural state. Fodor’s view has thus the radically implausible consequence that an othe-
rwise conceptless entity, e.g. a photographic plate, could grasp the concepto of justice. It has the
related consequence that one could grasp the concepto of justice without knowing anything
about the structure of that concepto and therefore without having any understanding of what it
is for one state of affairs to be more just than another.
In our view, one’s grasp of the concepto of justice is descriptive. The analysis that we are
proposing is obviously incompatible with Fodor’s.
 Chapter 24.  Peacocke on concept-possession 

Mental representations of properties are descriptive. But the descriptions in ques-

tion are not constructed (only) out of propositions or conceptso. They are (also) con-
structed out of representations out of concrete situations. Let us say that an analogue
mental representation of a situation is a “situation-representation.” (Perceptions and
memories thereof are examples of situation-representations.)
Given these points, suppose that Smith grasps the concepto triangle and that he also
grasps the concepto straight-edged figure. We need to account for the fact that Smith is
likely to believe an object to fall under the second concepto if he believes it to fall under
the first. Our analysis enables us to do this. For Smith to grasp either concepto is for him
to understand what various situations have in common. For x’s being a triangle to entail
its being straight-edged is for the one class of situations to be less inclusive than, but
otherwise coincident with, the other; and for Smith to be aware of this entailment is for
him to be aware of that inclusion-relation.5 Given what conceptso are (namely, com-
monalities among commonalities among....situations); and given what it is to grasp a
concepto (namely, to be aware of the existence of such commonalities); it follows auto-
matically that Smith is in a position to believe x is straight-edged if he believes x is a
triangle; and it also follows that Smith will believe the second, given a belief in the first,
so far as he is rational and so far as he is not inhibited by performance-related factors.
But, if our analysis is right, it would be false to say that Smith’s grasping the one
concepto consisted in his grasping the other. In virtue of grasping the first concepto, one
can grasp the second: but one needn’t do so. The kind of knowledge that is constitutive
of a grasp of the concepto straight-edged is constitutive of a grasp of the concepto trian-
gle. But a grasp of the concepto straight-edged is not itself constitutive of a grasp of the
concepto triangle.
This is exactly what we want. So far as one’s grasp of the concepto straight-edged is
constitutive of one’s grasp of the concepto triangle, one’s believing x is a triangle cannot
cause one to believe x is straight-edged. After all, if one’s grasp of the first concepto is
constituted by one’s grasp of the second, then thinking x is a triangle involves thinking
x is straight-edged: the first thought is the second thought – plus certain other thoughts
(e.g. x has three sides and x is closed). Since nothing can stand in any causal relation
with respect to itself or any part of itself, it follows that one’s thinking x is a triangle
cannot have a causal relation to one’s thinking x is straight-edged. Our analysis doesn’t
have this problem.
As we saw a moment ago, Peacocke’s analysis demands that your thinking x is a
triangle actually involve your thinking x is straight-edged and three-sided and closed.
This makes it hard to explain why propositions like triangles are three-sided, straight-
edged, closed figures can be informative. Our analysis doesn’t have this problem.
At the same time, our analysis accommodates the fact that, if you believe that x is
a triangle, you all but believe that x is straight-edged. According to our view, your be-
lief that x is a triangle is a conceptualization of a certain set of situation-representa-

5. Barwise and Perry (1999) argue for a similar position.

 Conceptual Atomism and the Computational Theory of Mind

tions, and your belief that x is straight-edged is a conceptualization of a certain set of

situation-representations that includes the first set. So if you believe that x is a triangle,
you either believe that x is straight-edged or you need only conceptualize what you
already know to generate that belief. This position is consistent with our intuition that
there is a constitutive relation between believing x is a triangle and x is straight-edged.
But it is also consistent with the fact that if x is a triangle, then x is straight-edged is not
as trivial as if x is an unmarried male, then x is unmarried.
Of course, as analytic statements go if x is a triangle is a triangle, then x is straight-
edged is fairly trivial. But obvious extensions of this line of thought show how there can
be analytic statements that are decidedly non-trivial – how there can be statements like
a circle is a closed two-dimensional figure of uniform curvature and a set S is infinitely
large exactly if its members can be put into a one-one correspondence with the members
of a proper subset of S.
These points give us some insight into the so-called “paradox of analysis.” 6 Some
background will help us identify this paradox. An adequate conceptualo analysis is
given by a true biconditional statement. An example would be:
(k) x is a circle iff x is a closed, two-dimensional figure of uniform curvature.
But not just any true biconditional will do. For example,
(k*) x is a circle iff x is a circle
is not an analysis. This is because it is trivial. And x is a unique contemporary U.S.
President iff x is George W. Bush is not a conceptualo analysis, even though it is non-
trivial and true. This is because it is not analytic. So an analysis is given by a true, ana-
lytic, and non-trivial biconditional; and a partial analysis (e.g. if x is a circle, then x has
a uniform curvature) is given by a non-trivial, analytic conditional.
Now we can state the paradox of analysis: How can a correct analysis be informa-
tive, given that the concepto on the left side of the biconditional must be identical with
the concepto on the right? If the concepto closed, two-dimensional figure of uniform
curvature is different from the concepto circle, then (k) fails to delineate the structure
of the latter, and thus fails as an analysis. But if those conceptso are identical, then (k)
should be as trivial as (k*).
What are we to say about this paradox? First of all, conceptualo analyses don’t state
conceptualo identities. The concepto circle is not identical with the concepto closed, two-
dimensional figure of uniform curvature. Those conceptso are different conceptualiza-
tions of the same situation-representations. Therefore, given a few obvious adjust-
ments, what we said in connection with the conceptso triangle and straight-edged
explains why (k) is both non-trivial and true.
If we don’t distinguish between conceptualo and non-conceptualo content, it is
quite impossible to explain how there can be such a thing as conceptualo analysis.

6. See Langford (1942).

 Chapter 24.  Peacocke on concept-possession 

Analyses expose internal, necessary relations between conceptso. But, as we just noted,
they don’t expose conceptualo identities. Given this, if we say that all content is concep-
tualo, then we must say that conceptualo analyses expose the “depths” or the “struc-
tures” of conceptso.
But that statement is easily shown to be false. The concepto class of all points equi-
distant from a given point in a plane is different from the concepto closed planar figure
of uniform curvature. This is equivalent to saying that the kind of thought that you have
in virtue of thinking that x falls under the one concepto is different from the kind of
thought you have in virtue of thinking that x falls under the other. In light of this, sup-
pose we say that conceptualo analyses (when true) expose the “depths” or the “struc-
tures” of conceptso. In that case, we are saying that when you think:
(i) x is the class of all points equidistant from a given point in a plane,
the content of your thought actually includes the proposition:
(ii) x is a closed planar figure of uniform curvature.
But when you think (i), you aren’t necessarily thinking (ii). When I think Sam is short
and Mary is tall, the proposition Sam is short is a part of the content of what I am
thinking. But (ii) is no part of the content of what I am thinking when I think (i).
One might say that, when I am thinking (i), I am “implicitly” thinking (ii). But if
taken as a psychological hypothesis, this is highly doubtful; and if taken as a non-psy-
chological, strictly logical thesis, it is just a misleading way of saying that (ii) is a cor-
rect analysis of what I think when I think (i).
If one holds that conceptualo content is the only kind, then the paradox of analysis
becomes unsolvable. But that paradox is quite solvable if we distinguish between con-
ceptualo and non-conceptualo content. A concepto is a conceptualization of content
that is not itself conceptualo; and an analysis of a concepto (a conceptualo analysis) is a
re-conceptualization of that same non-conceptualo content.
In this way we accommodate the fact that there is an internal connection between
the conceptso circle and closed figure. But we are not forced to say the one concepto has
the other as a veritable ingredient; and we are thus not forced to say that, when you
think circle, you necessarily think closed figure. At the same time, we have explained
why, if one grasps the first concepto, one has all but grasped the second, thereby ex-
plaining why conceptualo analysis produces a feeling of recognition – of seeing what
one has already seen, though from a new perspective – as opposed to a feeling of dis-
covery de novo.
 Conceptual Atomism and the Computational Theory of Mind

Conceptso versus properties

To facilitate the exposition of our argument, we have thus far assumed that conceptso
are identical with properties. According to this view, the concepto of triangularity is
identical with the corresponding property. This is not an unreasonable view, and some
philosophers hold it. But it is one that must at least be qualified. The concepto closed
three-sided, straight-edged, planar figure is obviously different from the concepto area
bounded by three straight-lines such that any two of them, but not all three of them, in-
tersect. But it is hard to see how those conceptso could correspond to different proper-
ties. Supposing that x falls under the concepto closed three-sided, straight-edged, planar
figure and y falls under the concepto area bounded by three straight-lines such that any
two of them, but not all three of them, intersect, there isn’t necessarily any difference
between x and y. In fact, x and y could be numerically identical. Where there are no
differences, there are no differences in properties. It thus seems that we have one prop-
erty (triangularity) and a multiplicity of conceptso. For similar reasons, the property
associated with the conceptO even prime is identical with the property associated with
the concept successor of one, even though those conceptso are clearly distinct. It seems
that to any one property there corresponds a multiplicity of conceptso.
This doesn’t mean that conceptso are not properties. But it means that the state-
ment conceptso are identical with properties must be qualified, lest it entail the false-
hood that the concepto even prime is identical with the concepto successor of one. A
point made by Frege (1954: 60–61) points us in the direction of what I believe to be a
correct understanding of the relationship between conceptso and properties. If you say
“x is a triangle”, you are ascribing a property to x. But for any individual x, the sentence
F
triangles are closed figures, but x is not a closed figureG is not self-contradictory. So
there is no individual x such that, if you say “triangles are closed figures”, you are saying
anything about x.
This suggests that statements like:
(T) triangles are closed figures
ascribe properties not to individuals, but to other properties. Of course, the property
of being a closed figure is a property of individuals, not of properties. After all, the
property of triangularity does not itself have the property of being a closed figure. The
property ascribed by (T) to the property of triangularity is not that of being a closed
figure, but is rather that of being such that any one of its instances is a closed figure.
Thus, conceptso are to properties what first-order properties are to individuals.
We saw in Chapter 21 that properties are commonalities that hold among common-
alities (that hold among commonalities…) that hold among situations. I thus propose
that conceptso are commonalities holding among the commonalities just mentioned. In
other words, conceptso are higher-order properties. Given that both (T) and
(C) circles are closed figures
 Chapter 24.  Peacocke on concept-possession 

are true, the property of being a circle has a property in common with the property of
being a triangle. They both have the property of being such that their instances are closed
figures. The concepto closed three-sided, straight-edged, planar figure obviously has a dif-
ferent composition from the concepto area bounded by three straight lines such that any
two of them, but not all three of them, intersect. That is why they are different conceptso.
But both conceptso correspond to a property that is unique to the property of triangular-
ity. The property of triangularity is uniquely such that all and only its instances are closed,
three-sided, straight-edged, planar figures; and the property of triangularity is also
uniquely such that all and only its instances are areas bounded by three lines any two of
which, but not all three of which, intersect. An individual falls under the one concepto if
it falls under the other. But this is not because that individual has a different property
corresponding to each of those different conceptso. It is rather because a property (that of
being triangular) had by that individual has a different property corresponding to each
of those different conceptso. We have thus made some case for the view, presupposed by
our analysis of analysis, that conceptso are properties.
 chapter 25

Semantics versus psychology

In this chapter, we will once again stop using the term “concepto” (and any derivatives
thereof, e.g. “conceptualo”), since context will make it clear whether or not we are dis-
cussing concepts in the platonic sense.
It seems reasonable to suppose that one must know the semantic rules of any lan-
guage that one knows and that linguistic ability is therefore posterior to cognitive abil-
ity. If this is right, then it is obviously viciously regressive to suppose that we think in a
language. Advocates of SCT deal with this by saying that one doesn’t have to know the
semantic rules of a language whose expressions mediate one’s own thoughts. It is nec-
essary, they say, only that one be “built to conform” to those rules.
In this chapter, we will see some reason (additional to those identified in earlier
chapters) to believe that, given a correct analysis of the concept of a semantic rule,
there can be no language where there is no knowledge of semantic rules and that SCT
is therefore guilty of vicious regressiveness.

The background

Why does “snow is white” mean snow is white and not grass is green? In general,
(EM) Why do our expressions mean what they do?
Grice (1957) had a theory about this. That theory is given by the statement that:
(G) Sentence S has meaning M exactly if people mean M by S; and for somebody
x to mean M by means of an utterance of S is for x to utter S with the intention
of being taken by the addressee to intend to convey M.
So for “snow is white” to mean snow is white is for people to mean the latter when ut-
tering the former. And for me to mean snow is white with an utterance of “snow is
white” is for me to produce that utterance with the intention of being taken by the
person to whom I am speaking to have the intention of conveying that snow is white.
There are several reasons to reject this view. First, many sentences that have never
been uttered, or even contemplated, have determinate meanings. This is obviously in-
compatible with Grice’s view.
Second, as Searle (1969: 142-150) and Blackburn (1984: 110-118) have conclu-
sively shown, one can mean M with an utterance of S even if one doesn’t intend, in
producing that utterance, to be taken to intend to convey M.
 Conceptual Atomism and the Computational Theory of Mind

Third, as those same authors also show, there are cases where one intends, in utter-
ing S, to be taken to convey M, even though one does not, in so doing, mean M. (At-
tempts have been made to fix up Grice’s analysis, so as to immunize it from this prob-
lem and the previous one. But in addition to having an ad hoc character, those attempts
are subject to innumerable counterexamples, as Searle and Blackburn make clear.)
Fourth, speaker’s meaning often diverges from literal meaning. Grice’s theory is
systemically incapable of registering this fact, since it identifies speaker’s meaning with
literal meaning.
Fifth, Grice’s theory circularly analyzes meaning in terms of itself. There are no
sentences until noises (or ink-marks or whatnot) are associated with meanings. So the
very concept of a sentence, and thus of a sentence-utterance, presupposes that of lin-
guistic meaning. Grice’s analysis is therefore circular, given that expressions such as
“symbol” and “sentence” enter into it. In an attempt to deal with this problem, one
might propose that the occurrences in (G) of “sentence” and “utterance” be replaced
with meaning-innocent terms, such as “noise” and “ink-mark.” But it is obvious that,
were these substitutions to be made, the resulting statement would be meaningless.
But we have yet to identify the most fundamental problem with Grice’s analysis.
(The point we are about to make was first made by Searle 1969: 42-50.) Unless I tbe-
lieve there to be a pre-existing symbolic relationship between “Socrates was wise” and
the proposition that Socrates was wise, I cannot mean the latter by an utterance of the
former. In general, a person cannot mean X by Y unless he believes that Y means X. (Of
course, as we discussed earlier, one doesn’t have to think that X has Y for its literal
meaning. One need only believe that X means Y in some way or other. The meaning-
relation in question could be one of non-semantic implicature.) Since the essence of
Grice’s theory is that linguistic meaning is derivative of speaker’s meaning, there is little
hope for Grice’s theory or for any elaboration of it. That is why few currently advocate
Grice’s theory of meaning.
But given only that the Gricean theory is incorrect, it doesn’t follow that semantic
relations are to be understood in non-psychological terms, and we saw in Chapter 20
that, indeed, they cannot be understood in this way.
I would now like to propose a third view. According to this view, semantic rela-
tions are not to be understood in terms of what people mean, but they are to be under-
stood in terms of what people think.

Some conditions that a semantic theory must satisfy

A semantic model cannot be viciously circular, i.e. it cannot itself use the notion of
meaning. Suppose that somebody put forth the following hypothesis:
(H) Tokens of “snow is white” mean snow is white because people agree among
themselves that “snow is white” is to mean snow is white.
 Chapter 25.  Semantics versus psychology 

At first, this view doesn’t seem particularly unreasonable. After all, you and I could
agree to let “Skippy” be our code-name for a friend of ours. But (H) will not do as a
general hypothesis as to the nature of expression-meaning. This is because (H) itself
uses the concept of expression-meaning, and it thus presupposes an antecedent under-
standing of that concept.
Permit me to clarify this. (H) has the form:
(*) Expression E has meaning M because people agree that E is to mean M.

Any statement having that form uses the concept of linguistic meaning. It would thus
be viciously circular to suppose that linguistic meaning could generally be understood
in terms of hypotheses of that form. (At the same time, linguistic rules can sometimes
be understood in terms of such hypotheses. As we said, you and I could agree that
“Skippy” was to be our code-word for a friend of ours.)
Let us extend this line of thought. Suppose that somebody put forth the following
hypothesis:
(HT) Tokens of “snow is white” are true iff snow is white because people agree
among themselves that “snow is white” is to be true iff snow is white.
Superficially, (HT) seems better than (HM). The concept of truth seems to be pre-lin-
guistic. A cat can truly believe that you are about to feed it. A dog can truly believe that
you are about to kick it. In general, non-linguistic creatures have true and false beliefs.
And, of course, they can also have accurate (true) and inaccurate (false) perceptions.
So true and false mental states are obviously antecedent to language.1

1. In any case, this is what I believe, and it is what most people (and most philosophers and
practically all contemporary psychologists) believe. Surprisingly, some authors have denied that
animals and pre-verbal infants have any true or false mental activity. See, for example, Davidson
(1980b: 141-170). Followers of the late Wittgenstein tend to have an attitude of antagonism to-
wards any theory that ascribes intentionality to non-communal or non-linguistic creatures. See,
for example, McDowell (1994) and Brandom (1998). We have already seen why no legitimate
basis for such antagonism is found in Wittgenstein’s work.
I should point out that Fodor (1981) has done a superb job of articulating why Wittgenstein
has failed to provide any legitimate reason to reject the basic tenets of cognitive science. An
implication of Fodor’s cogent discussion is that, indeed, non-linguistic creatures can think.
(More accurately, an implication of Fodor’s discussion is that creatures that don’t know any pu-
blic languages can think. Fodor, of course, believes that all thinking creatures ipso facto know
some kind of language. We’ve seen that Fodor is wrong about this, but this doesn’t change the
fact that, if we take the term “language” in a traditional sense, Fodor has established that one can
think without knowing a language.)
Donald Davidson (1980b: 141-170) held that, in order to think, a creature must know a
language; and he produced some interesting arguments in defense of this view.
But Davidson’s arguments presuppose an acceptance of a very extreme form of content-
externalism and must therefore be rejected, given what we saw in Part I of this work.
 Conceptual Atomism and the Computational Theory of Mind

Further, the notion of linguistic truth is plainly derivative of that of propositional

truth. The proposition there is water on Earth was true long before there were any
creatures or, consequently, any language.2 So given the fact that truth is not a distinc-
tively linguistic notion, it might seem that (HT) is not guilty of vicious circularity.
But this is a mistake, and (HT) is viciously circular. The notion of truth is, indeed,
not a distinctively linguistic notion. But the very question – or, at least, one of the very

There is another problem with Davidson’s view. If Davidson is right, then ex hypothesi one’s
linguistic competence isn’t mediated by any kind of intelligence. There are no cognitive proces-
ses mediating between perceptual input and verbal output.
In that case the speaker is not significantly different from Pavlov’s dog. But Chomsky (1959)
made it clear why Pavlovian (behaviorist) theories of linguistic behavior cannot be accepted.
Given Chomsky’s arguments, it is fair to say that linguistic competence can no more be explai-
ned without positing cognitive processes mediating between sensory input and verbal output
than Brownian motion can be explained without assuming the actuality of molecules. There is
thus good reason to believe that any theory like Davidson’s is a non-starter.
Further, Davidson’s position is incompatible with the very fact that anyone means anything
by anything. As we’ve seen, speaker’s meaning presupposes conventional meaning. I cannot
mean proposition P by my production of noise N unless I believe that N already means P. Since
Davidson’s position is that one must know a language in order to believe anything, it follows that
Davidson’s position is incompatible with the fact that anyone means anything by anything.
Knowledge of language obviously facilitates thought. But it cannot be constitutive of one’s
ability to think, since the ability to think is a prerequisite to learning a language. We have seen
independent grounds for this view in the present work. I should also point out that I try to pre-
sent a cogent argument for it in my article “Must one know a language to grasp propositions?”
(Kuczynski 2005b). In any case, it is not entirely unreasonable to assume that pre-linguistic crea-
tures can have true and false mental contents: and this is all that is demanded by the argument
presented in this chapter.
2. Frege (1884) eloquently defends this very point. Some philosophers deny that there are
propositions and would therefore, presumably, deny that there was anything to be true or false
prior to thought. But this seems straightforwardly false. Of course, nobody grasped that propo-
sition before there was thought. But that is irrelevant, unless it is assumed that propositions must
be grasped to exist. But in addition to begging the question, such a supposition would involve a
blatant non-sequitur, given that there is obviously no inference from X is grasped to X must be
grasped in order to exist. (Stars can exist without being grasped.)
In fact, it is incoherent to suppose that something must be grasped in order to exist. If so-
mething has no existence apart from awarenesses of it, then it collapses into those awarenesses,
in which case those awarenesses are not really awarenesses of anything. Admittedly, there do
seem to be some cases where a thing’s esse (its being) collapses into its concipi (its being thought
about). For, as Searle argues in the Social Construction of Reality, our legal and financial institu-
tions couldn’t exist unless somebody were aware of them. But it isn’t reasonable to suppose that
propositions are comparable in this respect to social institutions, given that those institutions
presuppose the existence of language and, therewith, of propositions. (Also, as Searle himself
makes clear, our social institutions are not ultimately counterexamples to our point that a thing’s
esse cannot collapse into its concipi. But a discussion of Searle’s demonstration of this would
take us too far off course.)
 Chapter 25.  Semantics versus psychology 

questions – that a semanticist must answer is “what is it for an expression to be true?”

The question “what is it for ‘snow is white’ to mean that snow is white?” is scarcely dif-
ferent from the question “what is it for ‘snow is white’ to be true iff snow is white?” So
even though truth is a pre-linguistic notion, (HT) is still viciously circular; for the very
question that we are trying to answer is, in effect, “what is linguistic truth, and what is
it for an expression to be true under such and such circumstances?”
The first thing anyone trying to answer (EM) must do is to find some notion such
that the concepts of expression-meaning (and expression-truth) can be non-circularly
understood in terms of that notion. There is such a notion: it is the notion of success.

The concept of linguistic success

Let us set aside language for the moment, and let us instead discuss the game of tennis.
The game of tennis is being played in a given context exactly if, in that context, certain
things count as success as other things count as failure. If you hit a certain ball that has
bounced no more than once on your side of the court into your opponent’s side of the
court and your opponent then fails to do the same thing (mutatis mutandis), you have
won a point, and you have thus had a success of a certain kind. (It is obviously irrele-
vant that, in tennis, points are measured in multiples of fifteen.) If you have a certain
number of successes of this kind before your opponent, then you have another kind of
success (you win a game). If you have a certain number of successes of this kind before
your opponent, then you have another kind of success (you win a set). And so on.
For a game of tennis to be played in a context C just is for it to be the case that, in
C, doing these things count as succeeding, and not doing them counts as failing. A
game of tennis is being played in C exactly if those success-conditions and failure-con-
ditions are operative, i.e. are instantiated, in C. So the game of tennis itself is a prop-
erty had by situations exactly if, in them, those conditions apply. And a particular
tennis-match, i.e. an instance of the game of tennis, consists in the instantiation of
those conditions.
Here we have to be careful. It is not as though the game of tennis exists first, and
these rules subsequently impose some kind of structure on this antecedently existing
entity. There is no such thing as tennis until it is the case that, in certain situations, one
enjoys a certain kind of success iff one hits a ball (that has bounced no more than once
in a certain place) over a certain kind of net and the person standing on the other side
of that net fails to do the same thing.
The concept of success, in this context, is thus not to be defined in terms of the
game of tennis. On the contrary, the game of tennis is to be defined in terms of that
concept, and that concept is to be taken as a primitive.3

3. Searle (1969: 37-47) makes very similar points.

 Conceptual Atomism and the Computational Theory of Mind

Of course, that concept is not primitive tout court. From a psychoanalyst’s per-
spective it may not be primitive and may represent, for example, a symbolic reenact-
ment of some primordial situation. But, for the reasons just given, the concept of suc-
cess, in the relevant sense, is primitive with respect to the concept of the game of tennis.
Here we must remember Searle’s (1969) distinction between “constitutive” and
“regulatory” rules. People bought and sold stock long before there was a law prohibiting
insider trading. Therefore that law isn’t constitutive of the practice of buying and selling
stocks; it merely regulates that practice. Here we are dealing with a “regulatory” rule.
But the rules discussed a moment ago are in a different category. It isn’t as though
people were playing tennis first, and those rules were subsequently imposed on this
activity. Rather, those rules constitute the game of tennis. As Searle (1969: 33–42)
points out, the rules of tennis, chess, and other games are rules in the constitutive, not
the regulatory, sense. Searle also says that semantic rules are rules in the constitutive,
not the regulatory, sense. I believe that Searle is correct, and I would now like to de-
velop his view; I would like to show that, on the basis of what Searle says, it is possible
to develop a viable answer to (EM).

Applying these points to semantics

For the sake of simplicity, let us initially consider only assertoric utterances.
There is some point of view from which an utterance of “Socrates was wise” is a
success just in case Socrates was wise, and from which, in general, an utterance of
FSocrates has phiG is a success iff Socrates has the property phi.
This doesn’t mean that if you say “Socrates was wise”, and he was in fact wise, any-
thing particularly desirable will happen to you. True utterances don’t necessarily bring
one wealth, fame, or happiness. But it does seem to be a simple fact that, from one
important perspective, an utterance of “Socrates was wise” is a success exactly if Socra-
tes was wise, and an utterance of “Socrates was not wise” is a failure under that same
circumstance.
Here we have to recall a distinction that we made when discussing Wittgenstein’s
Private Language argument: the distinction between linguistic success per se and the
consequences of such success. A detour through the game of tennis may once again be
useful. If you beat your wealthy and vain father-in-law at tennis, you have enjoyed a
kind of success (tennis-success). But the consequences of your tennis-success may be
disastrous: he disinherits you, relieves you of your sinecure at his firm, and so on.
Similarly, if you say “Socrates was wise”, you deserve credit for a kind of success
(linguistic success). But the consequences of that success may outweigh the success it-
self. Suppose that, for some reason, your boss hates anyone who thinks that Socrates
had any redeeming characteristics, and he therefore fires you when you say “Socrates
was wise.” Here your linguistic success was eclipsed by subsequent non-linguistic fail-
 Chapter 25.  Semantics versus psychology 

ures. And had you instead said “Socrates was not wise”, the consequences of your lin-
guistic failure might well have been outweighed by success of a non-linguistic kind.
Comments similar to those made a moment ago about tennis are true of language.
It isn’t as though the English language existed first and then there came into existence
various facts about success- and failure-allocation of the kind just described. It isn’t as
though the English-language existed first and then it came to be that one enjoyed success
if one uttered “Socrates was wise” iff Socrates was wise. More generally, it isn’t as though
the English language existed first and then it came to be that utterances of F...Socrates...G
met with success just in case Socrates had property…x… The English language didn’t
exist until it became the case that, given facts about how the world is, one enjoyed a
certain kind of success iff one fulfilled conditions of the kind just described.
(The statement that says what it is for “Socrates was wise” to mean Socrates was
wise is a multiple-biconditional or, more accurately, has such a biconditional embed-
ded within it. For that statement has the form: there is a certain kind of success S such
that an utterance of “Socrates was wise” is true iff Socrates was wise iff such an utterance
is met with S iff Socrates was wise. Given obvious extensions of what we just said in
connection with “Socrates was wise”, it follows that all semantic rules are given by
statements that have multiple bioconditionals embedded within them. For that reason,
the sentences that describe the basic structure of language are harder to understand
than those that describe the basic structure of tennis. But otherwise the two sets of
statements are comparable.)
So for English to be spoken in a given context just is for it to be the case that (inter
alia) an utterance in that context of Fx is redG is a success exactly if x is red, and an ut-
terance in that context of “Socrates was not wise” is a failure exactly if Socrates was wise.
The concept of success, in this domain, is not to be understood in terms of the English
language. On the contrary, the English-language is to be understood in terms of the
concept of that kind of success, and that concept is thus to be taken as a primitive in this
context. (Of course, that concept is not primitive from a psychological perspective.

Why the concept of language is not to be understood in terms

of the concept of truth

It is typically thought that the concept of language is to be understood in terms of the

concept of truth, not of success. There are two reasons why this is not the case. Any
attempt to understand linguistic meaning in terms of truth is bound to result in a
theory that is either viciously circular or that fails to distinguish actual from merely
possible semantic rules. As we saw in Chapter 20, we cannot explain the fact that “Soc-
rates snored” means Socrates snored by saying that the relevant semantic rule is a func-
tion that assigns truth to an utterance of “Socrates snored” exactly if Socrates snored.
After all, there is a function that assigns truth to an utterance of that same sentence
exactly if Socrates did not snore. This shows that the mere existence of a function of the
 Conceptual Atomism and the Computational Theory of Mind

kind described is insufficient for the existence of an operative semantic rule. (And a
“non-operative” semantic rule is no more an actual semantic rule than a possible indi-
vidual is an actual individual.) So a theory that identified semantic rules with func-
tions from utterances to truth-values fails to distinguish semantic rules from their pos-
sible but non-existent, counterparts.
In response, one might say, as Lewis (1975) does, that an operative semantic rule
is a function that people use. But then we must ask: what it is for people to use a func-
tion that assigns truth to utterances “Socrates snored” iff Socrates snored? It is presum-
ably for there to be an understanding among people that utterances of “Socrates
snored” are to be considered true iff Socrates snored. But, as we saw a moment ago,
such an understanding is itself to be understood in terms of the existence of semantic
rules. An agreement that utterances of “Socrates snored” is to be assigned truth iff
Socrates snored is not significantly different from an understanding that such utter-
ances are to mean Socrates snored. Thus, the concept of such anunderstanding is itself
to be understood in terms of the concept of a semantic rule, and the latter concept is
therefore not to be analyzed in terms of the former. So if we identify semantic rules
with functions from utterances to truth-values, our theory is either viciously circular
or it fails to distinguish existent from non-existent semantic rules.
There is a second reason why the concept of a semantic rule is not to be under-
stood in terms of the concept of truth. Many well-formed utterances are neither true
nor false. Because of this, their semantics cannot be understood in terms of the con-
cept of truth. But their semantics can be understood in terms of the concept of success.
If I ask “did Socrates snore?”, my utterance is a success, in the relevant sense, exactly if,
in response to it, the person I am addressing correctly affirms (or correctly denies) the
proposition that Socrates snored. If I say, “Socrates, snore!”, my utterance is a success
exactly if, in response to it, my auditor Socrates snores.
There are obviously appropriateness-conditions for questions and imperatives.4
The question “was Socrates wise?” isn’t true or false; it doesn’t have truth-conditions.
But it is obviously a meaningful utterance, and any viable semantics must accommo-
date that fact. It is obviously an important part of the semantics of “was Socrates wise?”
that the person to whom an utterance of that sentence is directed is, in some sense, sup-
posed to correctly affirm either that Socrates was wise or that Socrates was not wise. It is
not as though the sentence “was Socrates wise?” first had a semantics, and then there
arose this fact as to how auditors of tokens of it are to react to it. No – quite obviously,
this fact about how auditors are to react to utterances of it is constitutive of its semantics.
The semantics of questions – and (by obvious analogues of the argument just given)

4. The same is true of sentences having other moods, e.g. optatives (“would that she were
here!”) and exhortatives (“let’s go!”). But obvious adaptations of what we say about assertions,
questions, and orders apply to sentences of these other moods, and we therefore needn’t discuss
them explicitly.
 Chapter 25.  Semantics versus psychology 

imperatives, optatives, and so on – are to be understood in terms of the concept of suc-

cess. We will substantiate this point in the last section of the present chapter.

The meanings of sub-sentential expressions

The meanings of sub-sentential expressions, e.g. “Socrates”, are to be understood in

terms of their effects on the success-conditions of utterances in which they occur.5 As
we’ve discussed, an assertoric utterance of “Socrates was wise” is a success, in the rel-
evant sense, exactly if Socrates was wise. More generally, FSocrates has phiG (or, equiv-
alently, F…Socrates…G) is a success exactly if Socrates has phi (or, equivalently, exactly
if…Socrates…). Going up another level of generalization, E refers to O exactly if (as-
sertoric) utterances of the form F…E…G are true exactly if O has the property…x…
So even though an utterance of “Socrates” does not by itself have success-condi-
tions (leaving aside cases where such an utterance is elliptical for a whole sentence),
the semantics of “Socrates” is to be understood in terms of the success-conditions of
sentences of which it is a part.
In response to this, it might be said that “Socrates” is to be defined non-contextu-
ally – that the semantics of “Socrates” is given by the non-contextual definition: “Soc-
rates” refers to Socrates. But it would be meaningless to say that “Socrates” referred to
Socrates but then to deny that, in virtue of having the form FSocrates has phiG, a sen-
tence had anything to do with Socrates or, more precisely, with Socrates’ having phi.
When we say that “Socrates” refers to Socrates, we are saying that FSocrates has phiG is
true iff Socrates has phi, that an appropriate answer to Fdoes Socrates have phi? G is a
correct affirmation of either that Socrates has phi or that Socrates does not have phi
(depending on whether Socrates has phi or not), that an appropriate response to an
utterance of FSocrates, have phi! G consists in Socrates’ coming to have phi in conse-
quence of that utterance, and so on.6
Of course, not everyone will agree with every aspect of what was just said. But what
seems uncontroversial is that there is a certain kind of success S such that it is constitu-
tive of the semantics of “Socrates was wise” that assertoric utterances of it have S if
Socrates was wise, and lack S otherwise. More generally, it is clearly constitutive of the
semantics of assertoric sentence-tokens of the form FSocrates has phiG that any such
token is a success exactly if Socrates has phi. Similar remarks apply to questions of the
form Fdoes Socrates have phi?G and to imperatives of the form FSocrates, have phi! G
Supposing that M is the actual semantics of “Socrates”, it would be very artificial to
maintain that the word “Socrates” first had M and that, subsequently, the facts just
described (regarding the conditions under which uses of “Socrates” are successful)

5. Versions of this point are made by Tughendat (1970), Dummett (1973: 196-203) and also
– as both of those authors point out – in Frege (1884).
6. See Tughendat (1970) and Dummett (1973: 196-202).
 Conceptual Atomism and the Computational Theory of Mind

were superimposed on that semantic fact. It is much more natural to suppose that
“Socrates” has M in virtue of the fact that these success-conditions are operative.

The concept of linguistic success (continued)

Even though a token T of “was Socrates wise?” isn’t true or false, there is still some
proposition P such that T is a success, in the relevant sense, exactly if P is true. As we
discussed, T is a success exactly if the person to whom it is directed correctly affirms
either that Socrates was wise or its negation. So the verbiage to the right of the “exactly
if ” in the last sentence picks out the appropriate value of P. Obvious adaptations of
what we just said are true of all tokens of the form F…Socrates…G, and not just of to-
kens of that form that are in the interrogative mood. Of course, what we just said about
“Socrates” is true mutatis mutandis of all utterances.
This is not to say that all expressions refer. (In Chapter 22, we saw why, in light of
an argument of Frege’s, it is not possible to maintain that all expressions refer.) But it is
to say that, given any meaningful expression e, e’s semantics is given by some proposi-
tion to the effect that, for any utterance T of the form F…e…G, there is some proposi-
tion P such that T is a success exactly if P is true. No matter what views one has as to
the semantics of “and” or “or” or case- or tense-markers, no one will deny that the se-
mantics of such an expression is given in its entirety when it is said when it is appropri-
ate to use sentences containing them. Supposing (what may not be strictly true7, but
will serve as a convenient simplification) that a token of FP and QG is true iff P is true
and Q is true, then one knows everything there is to know about the semantics of
“and”, it being irrelevant whether one also thinks (as Montague did) that “and” refers
(to a function from pairs of sentences to truth-values) or whether one thinks (as the
early Wittgenstein did) that it does not refer to anything. If two people agree as to
whether the biconditional given a moment ago is correct, then they ipso facto agree as
to the semantics of “and” – even though they may disagree as to the nature of what it is
they are agreeing about. (Hilbert and Frege agreed that 7+5=12, even though they
disagreed as to the nature of what it is that they were thereby agreeing about.)
All of this may be distilled into the following statement:
(*) There is a certain kind of success S such that E is a symbol of a public language
(e.g. Urdu, English, Spanish) exactly if, given any token T of the form F…E…G,
there is some proposition P such that T has S exactly if P is true.

7. I say “may not be strictly true” because (to use the stock example) “John brushed his teeth and
went to the store” seems to have a different meaning from “John went to the store and brushed his
teeth.” It is a matter of controversy among semanticists whether the differences between those two
sentences are a matter of literal meaning.
 Chapter 25.  Semantics versus psychology 

For reasons already given, S is to be distinguished from monetary success, profes-

sional success, and any other garden-variety success.
Given that success and failure presuppose the existence of thought, it seems to fol-
low that the existence of a thought is indeed a prerequisite to that of anything having
any significant resemblance to a symbol of a public language.
This shows that if (*), or anything like it, is correct, then thought is prior to language,
and that SCT is in fact guilty of the vicious regressiveness of which it is often accused.

How to give a non-circular account of linguistic meaning

I would now like to indulge in what might at first seem like a fruitless and fanciful di-
gression. Strictly for the sake of argument, suppose that somebody proposed the fol-
lowing hypothesis:
(HB) Tokens of “Bob is at the store” mean Bob is at the store because there is an
agreement among people that, from a strictly linguistic viewpoint, such a to-
ken is a success exactly if Bob is at the store.
Before proceeding, I should make it clear what I do, and do not, hold. There is no
doubt that (HB) is false. (For expository reasons, let us momentarily forget about the
type-token distinction.) “Bob is at the store” is a complex expression. It means what it
does because of what its components mean; it means Bob is at the store because of what
is meant by “Bob”, “store”, “at”, and so on. So the reason that “Bob is at the store” means
Bob is at the store is not that people agreed to let that particular expression have that
particular meaning.
But, right now, that is irrelevant. Here what I wish to ask is this: Leaving aside the
question of whether it is true or even plausible, is (HB) circular? Is it characterized by
the same kind of vicious circularity as (HM) and (HT)?
My answer is “no.” But there are two reasons why many would say “yes.” Let us
consider these before proceeding. The first reason is that the expression “linguistic”
occurs in (HB).
So far as (HB) seems circular, that has nothing to do with its content, and is strict-
ly a matter of how it is formulated. First of all, there is no denying that some kind of
pre-verbal communication exists. When a dog growls at you, he is letting you know
that you should back off. And much of the information that human beings exchange is
non-verbal. How you respond to your boss isn’t a function only of what he is saying:
you take into account his facial expressions, body language, intonation, and so on. Of
course, it could be argued that it is only because he is verbally communicating with you
that your boss can also non-verbally do so. In other words, it could be held that his
non-verbal communiqués are parasitic on verbal ones. But this doesn’t appear to be
generally true of interpersonal non-verbal exchanges. Can’t pre-verbal toddlers com-
municate through cries and smiles? To be sure, we must take care to distinguish what
 Conceptual Atomism and the Computational Theory of Mind

we can infer from a toddler’s behavior (or countenance), on the one hand, and what the
toddler is trying to express, on the other. When a toddler is crying, we know that it is
distressed; but the toddler is not necessarily trying to communicate anything by cry-
ing. But in some cases that is precisely what the toddler is doing. (Parents quickly learn
to distinguish cases where the infant uses crying as a means of communication – as, for
example, a way of summoning the parent – from cases where the infant is simply react-
ing to its own distress and isn’t trying to communicate anything.) So there are surely
cases of bona fide non-linguistic communication that are not parasitic on linguistic
communication. Further, we would eviscerate the concept of language if we said that
all attempts to communicate were ipso facto linguistic. In general, therefore, language
isn’t required for people to exchange certain pieces of information. Rather, language
seems to be a way of extending the scope and precision of existing, non-verbal ways of
communicating.
As we just noted, the activity of exchanging information does exist among pre-ver-
bal and non-verbal creatures, albeit in an a relatively modest form. That activity defines
a unique field of endeavor: a sphere of conduct to which a special kind of success at-
taches. (Remember Aristotle’s point that, to each kind of endeavor, there corresponds a
unique kind of success.) Let us refer to that sphere of conduct as SC. Given this, we re-
move the aforementioned circularity from (HB) by rephrasing it thus:
(HB*) Tokens of “Bob is at the store” mean Bob is at the store because there is an
agreement among people that, within SC, such a token is a success exactly if
Bob is at the store.
Of course, (HB*) presupposes that our pre-linguistic ancestors – the ones who had no
language before they themselves instituted it – were aware of SC’s existence. In other
words, it presupposes that, at some level, they knew there to exist among themselves an
activity of exchanging information and that they had at least some correct views as to
the nature of that information-exchange. But this is obviously a reasonable supposi-
tion; for it is not out of the question to suppose that pre-linguistic humans were intel-
ligent enough to know that they had methods of communication and, therefore, to
suppose that they had at least some correct views about those methods. Thus (HB*) is
not guilty of the just-described vicious circularity.
But one might believe (HB), and also (HB*), to be guilty of another kind of circu-
larity. It has been said that the very instituting of agreements presupposes language. If
this is the case, then (HB) and (HB*) are indeed viciously circular.
But given the points made a moment ago, it is hard to take this objection seriously.
There is obviously non-linguistic communication. Obviously people sometimes make
themselves understood non-verbally. In response to this, it might be said that such
non-verbal understandings are parasitic on language. But we have already seen why
such a view is a non-starter: it is an obvious empirical fact non-verbal creatures can
make themselves understood to one another. Thus, taken as an empirical point, the
objection that all communication is linguistic is false.
 Chapter 25.  Semantics versus psychology 

As a counter-response, the objector might say that it is not an empirical, but an

analytic point, that any exchange of information is linguistic. But if that is right, then
any exchange of information is ipso facto linguistic, and the expression “linguistic
communication” becomes a pleonasm, like “unmarried bachelor.” We’d then have to
produce a new expression, e.g. “verbal behavior”, to distinguish between linguistic and
non-linguistic behavior. But then all of the questions that we currently have in connec-
tion with the term “linguistic behavior” would re-arise in connection with the term
“verbal behavior.” This shows that one is making a purely terminological point in say-
ing that exchanges of information are necessarily linguistic.
I will henceforth use the term “linguistic behavior” to refer to the activity of express-
ing oneself through speech, writing, and the like, and I will use the term “non-linguistic
behavior” to refer to the activity of expressing oneself in other ways (e.g. by kicking over
furniture, scowling, and so on).
There is one last possible problem with our analysis: it presupposes that proposi-
tions can be grasped independently of language. For there to be understandings of the
kind described by (HB) and (HB*), it must be possible to have a non-linguistic grasp
of the proposition that Bob is at the store. (Of course, non-linguistic entities wouldn’t
have stores. But that can be dealt with by changing the example.) Not all philosophers
would grant this. But there is good reason to think that such philosophers are wrong.
We’ve already seen that there is no linguistic activity where there isn’t a grasp of se-
mantic rules, and that a grasp of semantic rules is nothing other than a grasp of (meta-
linguistic) propositions. We’ve also seen that linguistic competence does not consist in
one’s having so many dispositions. Of course, given that I know that “Socrates” refers
to Socrates, I am disposed to use that expression in certain ways and not others. But
given what we saw at the end of Chapter 18, those dispositions are mere effects of my
linguistic competence, and are therefore not constitutive of it. In general, linguistic
competence must be seen as embodying bona fide knowledge – knowledge of proposi-
tions to the effect that such and such sounds (or ink-marks…) have such and such
meanings – and thus cannot be dispositionalized away. So just as our analysis requires,
the ability to grasp propositions is a prerequisite to the ability to use a language, and
the latter is therefore not constitutive of the former.

How to explain the meaningfulness of sentence-parts in terms

of the concept of success

I am proposing that it is agreements among people that endow noises and ink-marks with
linguistic meaning. (In this context, by an “agreement”, I mean an understanding – I obvi-
ously don’t mean a deliberate and conscious act of arriving at such an understanding.
Clearly, you and I didn’t decide to let “Socrates” refer to Socrates. But you and I have an
understanding to that effect.) Supposing this to be true, those agreements do not directly
 Conceptual Atomism and the Computational Theory of Mind

assign any meaning to whole sentences. They do so by way of agreements concerning the
expressions that compose sentences – expressions like “Bob”, “store”, “at”, and “wet.”
Here we confront an apparent problem for our analysis. Sub-sentential utterances
don’t have success-conditions. Leaving aside cases where single words are merely ellipti-
cal formulations of complete sentences, there is no circumstance in which an utterance
of “store” or “but” or “Fred” is a success (in the relevant sense). The concept of success, in
the relevant sense, is inapplicable to utterances of “Fred” or “store.”
The resolution of this difficulty is not hard to see, so long as we take into account
the fact that expressions are to be defined contextually. This point was made by Frege
(1884) and Russell (1905), and to solve our current problem we need only generalize
their insight.
As we saw in Chapter 3, tokens of “Fred” refer to Fred exactly if, in virtue of having
the form FFred has phiG, a sentence-token has for its meaning a proposition that is true
exactly if Fred has phi – exactly if, in other words, there is some x, such that x is identical
with Fred, and such that, in virtue of having the form FFred has phiG, a sentence-token
encodes the singular proposition: x has phi. Put another way, occurrences of “Fred” refer
to Fred exactly if, in virtue of having the form F…Fred…G, a sentence-token encodes a
proposition that has Fred as a constituent. (Right now we are going to consider only as-
sertoric sentences. We will soon generalize what we say about such utterances to ques-
tions, imperatives, and so on.)
In light of this, consider the following:
(F) There is an agreement among people to the effect that, in virtue of having the
form F…Fred….G, a noise (or ink-mark…) S encodes a piece of information P
such that Fred is a constituent of P and S is a success exactly if P is true.
(I said “noise (or ink-mark…)”, and not “expression”, because use of the term “expres-
sion” would have been viciously circular. For, as previously discussed, there can be
bona fide expressions only in a context where there is already language.)
What is the idea behind (F)? Consider the sentence “Fred is at the store.” As we’ve
discussed, there is some x such that x is identical with Fred and such that, from a
strictly semantic viewpoint, an utterance of that sentence is a success exactly if x is as
the store. Now consider the sentence “Fred is tall.” There is some x such that x is identi-
cal with Fred and such that an utterance of “Fred is tall” is a success, in the strictly
linguistic sense, exactly if x is tall. Such examples can be generated ad libidum. Given
these points, we may conclude that, in virtue of having the form, F…Fred…G, a sen-
tence S encodes a piece of information P such that P has the form…Fred...and such
that, in the strictly semantic sense, S is a success exactly if P is true.
So while utterances of “Fred” do not by themselves have success-conditions, whole
sentences of the form F…Fred…G do have such conditions; and the semantics of “Fred”
is given by identifying the success-conditions of sentences falling into this category. So
while it is true that no agreement among people directly gives meaning to “Fred is at
the store”, and while it is also true that sub-sentential expressions do not have success-
 Chapter 25.  Semantics versus psychology 

conditions, there is no reason to deny that sentence-level expressions are given mean-
ing through agreements among people as to the success-conditions of expressions.
“Fred” refers to Fred because of an agreement or understanding among people to the
effect that, if a noise (or ink-mark…) has the form F…Fred…G, there is some piece of
information P such that P is about Fred and such that S is a success, in the relevant
sense, exactly if P is true. (To echo what we said a moment ago: I said “noise (or ink-
mark…)” instead of “sentence-token” since, in this context, use of the latter would in-
volve vicious circularity.)
As we saw earlier, the relevant kind of success does not have to be understood in
linguistic terms, and the relevant kind of agreement does not presuppose the existence
of language. So in putting forth (F), we are not guilty of vicious circularity, and we are
able to account for the relevant facts about the semantics of “Fred.” In particular, we are
able to explain why “Fred” refers to Fred, i.e. why in virtue of having the form F…
Fred….G (or, equivalently, FFred has phiG) a sentence is true just in case…Fred…(or,
equivalently, Fred has phi).
If our analysis is right, then “Fred” is defined to be contextually, i.e. by giving the
truth-conditions of whole sentences containing it. This is not to say that “Fred” doesn’t
refer to Fred. Obviously it does. It is to say that for “Fred” to refer to Fred is simply for
sentences of the form F…Fred…G to have certain truth-conditions.
Independently of our analysis, this is obviously the case. As we said a moment ago,
it would be absurd to say that “Fred” referred to Fred, but to deny that, in virtue of be-
ing of the form FFred has phiG, a sentence was true iff Fred has phi. And it would be
absurd to deny that “Fred” referred to Fred but to maintain that, in virtue of being of
the form FFred has phiG, such sentences were true iff Fred had phi. So “Fred” refers to
Fred exactly if, in virtue of containing occurrences of “Fred”, sentences have certain
meanings or truth-conditions. Similar remarks apply to any referring expression.
Tughendat (1970) and Dummett (1973: 196–203) make similar points.
Obvious variants of what we have said about “Fred” apply to predicates (expres-
sions like “tall” or “wet”). There is some property PHI such that PHI=tallness and such
that, in virtue of having the form F…tall…G, a sentence-token S is such that for some
proposition P such that P concerns tallness (i.e. has that property as a constituent), S is
a success (in the previously discussed sense) exactly if P is true. Let S be a token of the
sentence “Tom is tall.” There is some piece of information or proposition P such that P
concerns (inter alia) the property of tallness such that, at the level of semantics, S is a
success exactly if P is true. (Of course, P is information identical or equivalent with the
proposition Tom is tall.) Given obvious extensions of these points, it is clear that all
so-called “lexical” items are to be defined contextually.
This isn’t to say that there is no such thing as denotative definition. It is to say that
denotative definition is a kind of contextual definition. If, for some object O, I say that
“Smith” denotes O, I am making two claims: first, that in virtue of having the form F...
Smith...G, a sentence-token encodes a proposition of the form...O...; and, second, that
“Smith” has certain syntactic properties. Let us elucidate the second point. In respect of
 Conceptual Atomism and the Computational Theory of Mind

syntax, “the property of wetness” is very different from “wet.” (For example, the one
can occur as a grammatical subject, whereas the other cannot.) Let P1...Pn be those
syntactic properties. In saying that “Smith” denotes x, I am saying (inter alia) that
“Smith” has P1...Pn or, at least, that it has some of those properties.

Why our analysis, unlike Grice’s, is consistent with the compositional character
of linguistic meaning

Grice’s view cannot be reconciled with the fact that what a complex expression means
is a function of what its parts mean.8 Obviously “Bob is at the store” means Bob is at the
store because each of its components has a certain meaning. Where sub-sentential ex-
pressions are concerned, speaker’s meaning necessarily diverges from expression-
meaning. Whatever it is that is meant by “at” or “store” or “Bob”, that meaning cannot
be the entirety of what some speaker means. (If you say “Bob” or “store”, then either
your statement is elliptical for some whole sentence, e.g. “Bob is at the store”, or you
cannot possibly mean what you are saying.) Therefore, any theory that, like Grice’s,
identifies expression-meaning with speaker’s meaning is straightforwardly incapable
of explaining what is meant by sub-sentential expressions.
Some Griceans counter-respond, extremely implausibly, by saying the following:
At some point in history, people uttered the noises socrateswaswise, socrateswast-
all, and socrateswasatthestore with the intention of affirming the propositions Soc-
rates was wise, Socrates was tall, and Socrates was at the store. In general, people
uttered whole sentences of the form F…Socrates…G with the intention of affirming
that…Socrates... Grice’s theory does apply to those sentences. After all, one can
certainly wish to affirm whole propositions about Socrates. (What you said a mo-
ment ago about “at” and “the” and other sub-sentential expressions is not true of
sentences.) What all of those sentences have in common is that they encode prop-

8. See Chomsky (1980: 83) and Platts (1979: 89-90). Chomsky and Platts are making what at
first appears to be a different point. They are not saying that Grice’s view is incompatible with the
compositional nature of sentence-meaning. Rather, they are saying that Grice’s view is incompa-
tible with the fact that sentences that have never been uttered already have a definite meaning,
given that nobody has ever meant anything by such a sentence. But these two points are equiva-
lent. Why do unuttered sentences already have fixed meanings? Because their constituents alrea-
dy have meanings, and there are recursive rules saying how those meanings fix the meanings of
sentences in which they occur. Where there is no semantic complexity, there is no possibility of
there being expressions that are never used but have fixed meanings. And where there is such
complexity, such expressions can exist. So the point of the objection that Chomsky and Platts are
making to Grice’s theory is equivalent with the objection that I am making to it.
Blackburn (1984: 127-130) discusses this objection, and says that it is null and void. We are
now going to consider Blackburn’s explanation as to why Grice’s theory is consistent with the
compositional character of meaning.
 Chapter 25.  Semantics versus psychology 

ositions of the form…Socrates…So the mechanism described by Grice makes it a

rule that socratessnored means that Socrates snored, that socrateswaswise means
that Socrates was wise, and so on. Thus, the Gricean story explains why sentences
having the form F…Socrates…G mean…Socrates… (Equivalently, the Gricean
story explains why sentences having the form FSocrates has phiG are true exactly if
Socrates has phi.) For “Socrates” to refer to Socrates just is for it to be the case that,
in virtue of having the form F…Socrates…G, an utterance is true exactly if…Soc-
rates…Therefore, contrary to what you argued, Grice’s theory does explain what it
is for “Socrates” to mean what it currently does. These points (mutatis mutandis)
are true of any given morpheme.9

Even if this hypothesis is correct, it shows only that historical antecedents of contem-
porary semantic facts can be understood in Gricean terms. It does not show that the
same is true of those facts themselves.
Permit me to clarify this. The hypothesis just stated has the form:
Such and such events led to its being the case that socratessnored, socrateswaswise,
and socrateswastall came to mean Socrates snored, Socrates was wise, and Socrates
was tall. And that fact, in its turn, led to its being the case that, in virtue of having
the form F…Socrates…G, an utterance is true exactly if…Socrates…

We want to know what it is for it to be the case that, in virtue of having the form F…
Socrates…G, an utterance is true exactly if…Socrates…If correct, the hypothesis under
examination tells us what led to that semantic fact. So it gives us a causal, not a concep-
tual, analysis of that fact. But we don’t want to know how that fact came about. We
want to know what that fact is: we want to know its structure. And the hypothesis in
question is silent on that score.
The following objection might be made in response to this point:
But surely that hypothesis does give us a conceptual analysis of that fact. After all,
that analysis tells us what it is for noises of the form…Socrates…to be true exactly
if…Socrates….For it tells us that noises of the form…Socrates…are true iff…Socra-
tes…exactly if such noises are produced by people with certain intentions. So that
explanation shows how the fact that such noises have such meanings constitutively
depends on people’s producing those noises with certain intentions. So in direct op-
position to what you say, that analysis tells us what it is for such noises to bear such
meanings. It doesn’t merely tell us what historical events led to that semantic fact.

This objection muddles a subtle but crucial distinction. Given only that socratessnored
means that Socrates snored, that socrateswastall means that Socrates was tall, and so
on, it doesn’t follow that it is in virtue of having the form…socrates…that a noise makes
a statement about Socrates. In other words, it doesn’t follow that a noise’s making a
statement about Socrates depends on its containing the sound socrates. Of course, it

9. This is Blackburn’s (1984: 129) position.

 Conceptual Atomism and the Computational Theory of Mind

follows that sounds of the form…socrates…have meanings that concern Socrates, but
not that it is in virtue of having that form that those sounds have those meanings.
Let L be a language that has three primitive symbols – namely, “socratessnored”,
“socrateswastall”, and “socrateswaswise” – whose meanings are, respectively, Socrates
snored, Socrates snored, and Socrates was wise. Because they are semantically primitive,
those sentences don’t have a semantic unit in common corresponding to the sound
socrates. So it is not in virtue of the fact that they all contain that sound that they are all
about Socrates. Of course, those sentences are semantically similar: they are all about
Socrates. And they are also acoustically similar: they all contain the noise socrates. But
it is not in virtue of their having that acoustical similarity that they have that semantic
similarity. Their being similar in both respects is only a coincidence. (That is, it is a
coincidence from a semantic point of view, though it may not be a coincidence from
some other point of view.)
If correct, the hypothesis under examination shows us that Grice’s story is compat-
ible with the fact that socrateswaswise means that Socrates was wise. But that hypoth-
esis doesn’t show that the Gricean story is consistent with the one fact that is relevant
in this context, namely: it is in virtue of the fact that it contains the sound socrates that
a present-day utterance of “Socrates was wise” makes a statement about Socrates. The
hypothesis in question gives us the semantics of L, not of English. It gives us the se-
mantics of a language all of whose expressions have English homonyms, but that isn’t
English, the reason being that its sentences have different decompositions from their
English counterparts.
The word “why” has both causal and conceptual meanings. If I want to know why
the interior angles of some surface add up to approximately 180°, I can be given either a
conceptual explanation (“because that surface is triangular, and the interior angles of a
triangle add up to 180°”), or I can be given a causal explanation (“because it used to be a
square, but the carpenter sawed it into two, triangular halves…”). If correct, the hypoth-
esis under examination provides a causal-historical explanation of the fact that the noise
socratessnored means Socrates snored. So in the causal sense of “why”, it tells us why that
noise has that meaning. But, regardless of whether it is correct, it doesn’t tell us what it is
for “Socrates” to mean what it currently does, or therefore how the meaning of a sentence
of the form F…Socrates…G depends on the meaning of “Socrates.”
The hypothesis in question exposes a problem that is endemic to Grice’s theory.
For the sake of argument, suppose that people uttered socrateswaswise, socratessnored,
and so forth, with the intention of affirming Socrates was wise, Socrates snored, and so
on. Given only that people did this, it doesn’t follow that it ever became a rule that one
must produce a sound of the form…socrates…to affirm a proposition about Socrates.
Given only that an activity is customary, it doesn’t follow that it is de rigueur. Of course,
norms often develop out of regularities. It is now a law in the U.S. that one must drive
on the right, and it is possible that this law developed out of what was once a mere
 Chapter 25.  Semantics versus psychology 

regularity.10 Similarly, it is theoretically possible that it was once merely customary for
people to produce sounds of the form…socrates…with the intention of affirming state-
ments about Socrates., and also that the rules now governing the use of “Socrates”
grew out of those regularities. But even if they are correct, these hypotheses merely
identify the historical precursors of the fact that “Socrates” now has a certain seman-
tics. Those hypotheses don’t constitute an analysis of what it is for “Socrates” to mean
what it currently does or, consequently, of how the meanings of sentences of the form
F…Socrates…G depend on the meaning of “Socrates.”11
Wittgenstein (1958) and Searle (1969) rightly emphasize that there is no language
where there are no rules: a noise or ink-mark doesn’t token a linguistic expression un-
less it represents an attempt on somebody’s part to follow an existing body of rules.
Grice tries to show that expression-meaning can be understood in terms of what peo-
ple are trying to affirm by means of various sounds (or ink-marks or hand-move-
ments…). If Grice’s account is to be free of vicious circularity, it must be assumed that
the sounds in question do not represent the outcomes of people trying to follow exist-
ing linguistic rules. But if a given person isn’t following existing rules, then what he
does is, at most, a simulation of actual linguistic behavior, it being irrelevant what
proposition that person is trying to affirm and also what noise or ink-mark that person
makes in his effort to affirm it. Therefore, the activity of such a person cannot be con-
stitutive of any semantic system, even though it is theoretically possible that such ac-

10. This example is borrowed from Blackburn (1984: 119). Blackburn (1984: Chapter 4) provi-
des an illuminating discussion of the problems relating to how mere verbal regularities could
give rise to rules governing word-usage.
11. As a matter of historical fact, the English word “Socrates” did not come into existence as a
result of people producing sounds of the form…socrates…with the intention of making state-
ments about Socrates. English-speakers didn’t go around producing such noises until “Socrates”
(or some other similar word) already referred to Socrates. That English word was borrowed
from a word of another language (Medieval French, we may suppose). In its turn, that word was
derived from a word of another language (Latin, we may suppose). And that word was derived
from a word of yet another language (Ancient Greek, we may suppose).
Obviously the Ancient Greek word for Socrates didn’t come about in a way that validates
the hypothesis under consideration. Suppose that sukrat is the sound of the Ancient Greek word
in question. It isn’t as though the Ancient Greeks first went around uttering noises of the form…
sukrat…and their word for Socrates eventually arose out of those pre-linguistic noises. No soo-
ner was Socrates given a name than that name already had the semantic and recursive properties
associated with “Socrates” and with any other well-formed expression of an already completely
developed language.
So the hypothesis in question obviously doesn’t apply to “Socrates” or, for exactly similar
reasons, to any expression constitutive of any language that we know of. If that hypothesis ap-
plies to anything, it is to primordial events that predate recorded history. This underscores the
fact that, if correct, Grice’s theory only identifies the historical precursors of semantic facts and
thus fails to analyze those facts themselves. It also shows how improbable it is that Grice’s theory
is correct even as a hypothesis as to what those precursors are.
 Conceptual Atomism and the Computational Theory of Mind

tivity might help bring such a system into existence. We see once again that Grice’s
theory involves a failure to distinguish conceptual from causal explanation.
Let us sum up this lengthy discussion. The essence of Grice’s theory is that expres-
sion-meaning is fixed by speaker’s meaning. But there are two reasons (additional to
the five identified earlier) why this theory is false. First, speaker’s meaning is itself a
relation towards (actual or perceived) expression-meaning. Second, speaker’s meaning
cannot possibly coincide with the meaning of sub-sentential expressions: what a
speaker means can never be what is meant by “at” or “the” or “Bill.” Grice’s theory can
explain, at most, what it is for “billisatthestore” to mean Bill is at the store – i.e. that
theory may be able to explain what it is for some homonym of “Bill is at the store” to
mean Bill is at the store – but it has nothing to tell us about that sentence itself. We may
conclude that, just as we alleged, Grice’s theory cannot be reconciled with the fact that
what complex expressions mean is a function of what their parts mean.
Our analysis does not have this problem. Built into our theory is an explanation as
to how the meaning of “Socrates snored” is derived from the meanings of its constitu-
ents. In fact, our analysis doesn’t have any of the problems associated with Grice’s su-
perficially plausible but systemically flawed identification of expression-meaning with
speaker’s meaning.
At the same time, our analysis is also consistent with the kernel of truth in Grice’s
theory, viz. that what people think is at least partially determinative of what words mean.
(Of course, what this or that specific person thinks may not be determinative of what any
English word means. But surely the fact that “red” means red is not entirely independent
of what everybody throughout time has thought.)
Further, our analysis isn’t guilty of vicious circularity. Non-linguistic creatures are
capable of having understandings or agreements with one another, and the specific
agreement described by (H) doesn’t involve any distinctively linguistic conceptso. To
be sure, it demands that non-linguistic creatures have the conceptso of truth and falsity
and, more generally, of information. So it presupposes that pre-verbal humans can
have some understanding of what it is to be misinformed or correctly informed. But
not only is this presupposition clearly consistent with the empirical facts: it is also ana-
lytically true. A language is a set of semantic rules, and grasping a language therefore
involves grasping such rules. Because such rules assign truth-conditions to expres-
sions, some understanding of what it is for a representation to be true or false is a
prerequisite to, and must therefore exist independently of, a grasp of any semantic
rules and, therefore, of any language.
Finally, our analysis is consistent with the fact that linguistic activity is necessarily
rule-governed. According to our analysis, E’s referring to O consists in its being the
case that there is a rule according to which utterances of the form…E…are successes
or failures, depending on whether…O…
 Chapter 25.  Semantics versus psychology 

Some additional, apparent problems with our analysis

The relationship between “Fred” and Fred is not strictly parallel to the relationship
between “wet” and the property of wetness. This point, or one like it, was made by
Frege. The expression “the property of wetness” refers to the property of wetness. But
“I sat on a wet chair” is meaningful and potentially true, whereas “I sat on a the prop-
erty of wetness chair” is not meaningful or, therefore, potentially true. Our analysis, it
might be said, models all semantic relations on the relationship between “Fred” and
Fred and is thus in error.
This objection is misguided. Our analysis is not to the effect that all semantic rela-
tions are to be understood in terms of the relationship between a singular term and its
referent. I chose to illustrate my theory with discussions of singular terms. (For obvi-
ous reasons, that theory would have been much harder to take in if my initial illustra-
tions had concerned adverbs or quantifiers.) The essence of our analysis is that seman-
tic rules can be understood in terms of agreements to the effect that sentences having
a certain property (e.g. the property of comprising “Fred” or “tall” or “or”) are ipso
facto to have certain success-conditions. Obviously the specific agreements involved
will vary from expression to expression and also from expression-category to expres-
sion-category. But there is no difficulty extending what we said about singular terms to
other kinds of expressions.
In fact, our theory is especially well-suited to giving the semantics of expressions oth-
er than singular terms. Given any expression that is not a singular term, it is obvious that
its meaning can be given only contextually. Conceivably, one might deny this, saying that
“wet”, “tall”, and “intelligent” refers to wetness, tallness, and intelligence. But that definition
would be inadequate since it wouldn’t distinguish (for example) “wet” from “the property
of wetness” – since, in other words, it wouldn’t provide the relevant syntactical informa-
tion. So while it is true that our analysis demands that expressions be definable contextu-
ally, it is clear on independent grounds that expressions other than singular terms must be
so defined; and we have seen that singular terms themselves must be so defined.
In fact, a slight generalization of a point made in Chapter 1 provides an additional
reason to hold that singular terms must be defined contextually. As we just saw, a defi-
nition is adequate only if it provides the syntactical properties of the expression in
question. You are missing crucial semantic information if you know what an expres-
sion refers to, but don’t know how to combine it with other expressions. So far as state-
ments like “that guy [accompanied by an ostension of person x] is named ‘Socrates’” or
“‘Socrates’ refers to Socrates” are adequate, it is because they are equivalent with state-
ments that give the syntactical properties of “Socrates.” Supposing that x is the person
ostended, an utterance of “that guy is named ‘Socrates’” is to the effect that any sen-
tence-token of the form F…Socrates…G means…x…(i.e. such an utterance it to the
effect that FSocrates has phiG is true iff x has phi). Since, ultimately, the only adequate
definitions are contextual definitions, it is not to the discredit of our theory that it de-
mands that expressions be contextually definable.
 Conceptual Atomism and the Computational Theory of Mind

Some additional problems with our theory (continued):

non-assertoric utterances

A more substantive problem with our analysis has to do with the fact that it doesn’t
apply to non-indicative sentences. But this problem is easily remedied. Non-assertoric
utterances are, of course, neither true nor false. But, as we noted, given any such utter-
ance S, there is some proposition P such that S is a success, in the relevant sense, ex-
actly if P is true. Suppose I ask you whether Smith went to the store. From a strictly
linguistic standpoint, my speech-act is a success exactly if, upon hearing it (or other-
wise encountering it), you correctly affirm either that Smith went to the store or its
negation.12 (For expository reasons, either pretend that you are Jones or substitute your
actual name for each occurrence of “Jones” in the present discussion.) So there is some
proposition of the form…Smith…such that my utterance is a success exactly if that
proposition is true. That proposition is not: Smith went to the store. Rather, that propo-
sition is: Jones correctly affirms whether or not Smith went to the store.
Suppose that I order you to meet Smith at the store. From a strictly linguistic
standpoint, my speech-act is a success exactly if, in response to hearing it, you make it
the case that you meet Smith at the store. So there is some proposition of the form…
Smith…such that my utterance is a success exactly if that proposition is true. But that
proposition is not: Jones meets Smith at the store. Rather, that proposition is: Jones
makes it the case that he meets Smith at the store.13
Here is what this suggests. “Socrates” refers to Socrates exactly if, given an utter-
ance S that has the form F…Socrates…G, there is in virtue of that fact some proposition
P having the form…Socrates…such that S is a success, in the relevant sense, exactly if
P is true. The semantic rule that pairs off Socrates with “Socrates” is thus an under-
standing among people that assigns such success-conditions to utterances of that form.
More generally, E refers to O exactly if, given an utterance S of the form F...E...G , there
is some proposition P such that P has the form...O...and such that S is a success exactly
if P is true. Even more generally, E has meaning M exactly if, given an utterance S of

12. Strictly speaking, there is some proposition P such that my utterance is a success exactly if,
in response to that utterance, you affirm either P or its negation. Suppose that you truthfully say
either “Smith went to the store” or “Smith did not go to the store”, but that you were not prompted
to do so by my utterance. For example, suppose that, as I am watching you give a speech on T.V.,
I foolishly ask you whether or not Smith went to the store and that, by sheer coincidence, you
truly say “Smith went to the store” right after I asked the question. Under that circumstance, it
seems to me, my speech-act was not a success, even though it was in some respects indistinguis-
hable from one that is a success. But in the present context, it isn’t necessary that we deal with this
nicety or with others like it. In Kuczynski (2005c), I give definitions of the terms “sentence”, “as-
sertion”, “question”, and “imperative” that are intended to be correct at the level of detail.
What we just said about questions is true mutatis mutandis of what we are about to say
about imperatives.
13. See the last sentence of the previous footnote.
 Chapter 25.  Semantics versus psychology 

the form F...E...G , there is some proposition P such that P has M as a constituent and
such that S is a success exactly if P is true.
A book-length defense of this analysis is given in Kuczynski (2005c). What is im-
portant here is that, if this line of thought is even approximately correct, it follows that
SCT is viciously circular, the reason being that the understandings mentioned in the
last paragraph obviously presuppose the existence of thought and, indeed, of proposi-
tional thought.
An advocate of SCT might counter-respond by saying that the representations
mediating our thought are not assigned their meanings by rules of that kind. But in
that case, they are not assigned meaning by anything comparable to the semantic rules
that assign meaning to the expressions of a public language, and it is thus no longer
appropriate to describe those representations as “sentences.”

The definition of “linguistic symbol”

In Chapter 14, we saw that all symbols are relations between morphological and psy-
chosocial factors, and that symbol-tokens are instances of such relations. But, as we
also saw, not all such relations are symbols. Given the points made in this chapter, we
are able to identify some of the other conditions that such a relation must satisfy to
qualify as a symbol.
First of all, only instances of morphology can be perceived. For this reason, any
assignment of meaning to a certain type of object must proceed by way of an indica-
tion of some instance of that type. Suppose that you and I want to be able to make
statements about Larry in his presence, but without his knowing that we are doing so.
We thus agree to let “Arnold” be our code word for Larry. This agreement, we may sup-
pose, is enacted as a result of my saying to you:
(*) Henceforth, “Arnold” will be our code-word for Larry.

Let ARN refer to the specific token of “Arnold” that occurred in (*). (*) assigns mean-
ing to a certain type of object. But what occurs in (*) is some instance of that type,
namely ARN. So (*) is to the effect that tokens similar in some respect to ARN refer to
Larry. There is thus some specific similarity-relation R such that my proposal is to the
effect that tokens bearing R to ARN refer to Larry.
For reasons that we’ve discussed in this chapter, there is a specific kind of success
S such that “Arnold” refers to Larry because the following holds: there is some object
O (namely, Larry) such that, in virtue of having the form F…Arnold…G, a sentence-
token has S iff…O…
Given this last point, along with the points made just before it, the following holds:
(**) There is a specific kind of success S, such that, for some object O (namely, Larry),
and some relation R, anything bearing R to ARN is a symbol-token (in our idiolect)
 Conceptual Atomism and the Computational Theory of Mind

denoting Larry, the reason being that you and I have an understanding to the effect
that, if m is an instance of morphology bearing R to ARN, then in virtue of that fact
F…m…G has S iff…O…

(**) makes it clear that, if a bit of morphology M refers to Larry, that is entirely in vir-
tue of the fact that we have an understanding concerning that bit of morphology (or,
more accurately, concerning the kind of morphology of which that particular bit is an
instance obviously our initial agreement didn’t concern each individual instance of the
relevant morphology-type). So (**) validates our suggestion that symbol-tokens are
instances of relations between psychosocial and morphological factors.
(**) can be generalized so as to yield an analysis of the concept of a linguistic sym-
bol and, therewith, of the concept of linguistic meaning:
(***) There is a certain kind of success S such that a bit of morphology m tokens a
linguistic symbol-type just in case, for some meaning (e.g. some individual or
property) M, at least some utterances of the form F…m…G have S just in
case…M…
An utterance of the form F…m…G has S just in case…M… just in case the
following two conditions are met. First, there is an understanding among peo-
ple to the effect that, for some bit (or, more likely, various bits) of morphology
B, there is some relation R such that, if a bit of morphology B* bears R to B,
then in virtue of having the form F…B*…G, an utterance has S exactly if…
M…. Second, m bears R to B.
(***) identifies some of the conditions that a relation between psychosocial and mor-
phological factors must satisfy to qualify as a linguistic symbol.
Of course, for this definition of the term “linguistic symbol” to be fully adequate,
S would have to be described more precisely than we have described it. But this is a
topic for another work.
 Conclusion

Semantics isn’t pre-semantics. Content isn’t truth-maker. Data isn’t meta-data.

In Part I, we found that, given these three distinctions, there are many reasons to
reject content-externalism and many reasons to accept semantic externalism.
Content externalism has some implausible consequences. It entails that one can
rationally accept both a proposition and its negation; it strips the mental of causal pow-
ers; and it makes one’s knowledge of one’s own conscious thoughts as uncertain as one’s
knowledge of the external world. We found that there is no reason to accept any of these
highly revisionist views, since the relevant data can be accommodated without them.
Advocates of content-externalism have attempted to reconcile that doctrine with
our pre-theoretic views about self-knowledge and mental causation. (Astonishingly,
they have generally accepted that one can rationally accept contradictions.) But those
attempts are failures.
We found that Kripke’s argument for synthetic necessary truth involves a failure to
distinguish between semantics and pre-semantics. We also found that a descriptivist
conception of conception can be reconciled with the fact that, in at least some cases, hav-
ing a concept of an object involves having a certain kind of causal connection to it.
Our descriptivist conception of conception is incompatible with Fodor’s concep-
tual atomism. But we found in Part II that atomism is false. Since atomism is false, so
is SCT. Since SCT is false, so is CTM.
We also found that, setting aside the erroneousness of conceptual atomism, there
are good reasons to reject SCT and CTM. Two are of special note.
First, in the arguments on behalf of CTM, the term “syntax” has at least three differ-
ent meanings. Sometimes it refers to an expression’s morphology; sometimes it refers to
its semantic decomposition; and sometimes it refers to its proof-theoretic relation to the
semantic rules constituting the language of which it is a part. The expressions “form”,
“algorithm”, and “mechanical procedure” are used in a comparably ambiguous manner.
Many of the arguments for CTM collapse once these ambiguities are recognized.
Second, supposing that Gareth Evans is right to distinguish between conceptual and
non-conceptual content, it follows that not all mental content has a sentential structure
and therefore that SCT and, consequently, CTM are false. We found that Evans’ distinc-
tion is both defensible and theoretically fruitful.
Content-externalism demands an acceptance of conceptual atomism and also of a
strictly causal theory of conception. We have found that all three of these doctrines are
false and that the relevant data can be modeled without them. Our model has the vir-
 Conceptual Atomism and the Computational Theory of Mind

tue of being entirely consistent with Kripke’s powerfully argued semantic insights. We
may conclude that both content-internalism and semantic-externalism are correct.
 Bibliography

Armstrong, David. 1989. Universals: an opinionated introduction. Boulder: Westview Press.

Ayer, Alfred Jules. 1952. Language, Truth, and Logic. New York: Dover Publications.
Ayer, Alfred Jules. 1954. “Can there be a private language?” In The Philosophy of Language, A.P.
Martinich (ed), 449–456. New York: Oxford University Press.
Bach, Kent. 1984. Thought and Reference. Oxford: Clarendon.
Barwise, Jon. 1989. The Situation in Logic. Palo Alto: CSLI Publications.
Barwise, Jon and Perry, John. 1983. “Semantic Innocence and Uncompromising Situations.” In The
Philosophy of Language, A.P. Martinich (ed), 392–404. Oxford: Oxford University Press.
Barwise, Jon and Perry, John. 1999. Situations and Attitudes. Palo Alto: CSLI Publications.
Berkeley, George. 1934. An Essay Towards a New Theory of Vision. Oxford: Clarendon Press.
Bilgrami, Akeel. 1992. “Can externalism be reconciled with self-knowledge?” In The Twin-Earth
Chronicles, A. Pessin and S. Goldberg (eds), 323–341. Armonk, NY: M.E. Sharpe.
Blackburn, Simon. 1984. Spreading the Word. Oxford: Clarendon.
Boguslawski, Andrzej. 1994. “Sentential Complementation and Truth.” In The Syntax of Sentence
and Text, S. Čmejrková and F. Štícha (eds), 317–332. Philadelphia: John Benjamins.
Bonjour, Laurence. 1985. The Structure of Empirical Knowledge. Cambridge: Harvard University
Press.
Bonjour, Laurence. 1998 In Defense of Pure Reason. Cambridge: Cambridge University Press.
Brandom, Robert. 1998. Making it Explicit. Cambridge: Harvard University Press.
Burge, Tyler. 1979. “Individualism and the Mental.” In The Twin-Earth Chronicles, A. Pessin and
S. Goldberg (eds), 125–141. Armonk, NY: M.E. Sharpe.
Burge, Tyler. 1982. “Other Bodies”. In The Twin-Earth Chronicles, A. Pessin and S. Goldberg
(eds), 142–160. Armonk, NY: M.E. Sharpe.
Burge, Tyler. 1986. “Individualism and Self-knowledge”. In The Twin-Earth Chronicles, A. Pessin
and S. Goldberg (eds), 342–354. Armonk, NY: M.E. Sharpe.
Cain, M.J. 2002. Fodor: Language, Mind and Philosophy. Cambridge: Polity Press.
Carnap, Rudolph. 1934. The Unity of Science. London: K. Paul, Trench, Kubner & Co, Ltd.
Carnap, Rudolph. 1937. The Logical Syntax of Language. London: Routledge & Kegan Paul.
Carnap, Rudolph. 1947. Meaning and Necessity. Chicago: University of Chicago Press.
Carnap, Rudolph. 1950. Logical Foundations of Probability Theory. Chicago: University of Chi-
cago Press.
Carnap, Rudolph. 1956. Introduction to Symbolic Logic and its Applications. New York: Dover.
Carnap, Rudolph. 1966. Introduction to the Philosophy of Science. New York: Dover.
Chalmers, David. 1996. The Conscious Mind. New York: Oxford University Press.
Chomsky, Noam. 1959. “A Review of B. F. Skinner’s Verbal Behavior.” In Language 35 (1): 26–58.
Chomsky, Noam. 1965. Aspects of the Theory of Syntax. Cambridge: M.I.T. Press.
Chomsky, Noam. 1975. Reflections on Language. New York: St. Martin’s Press.
Chomsky, Noam. 1980. Rules and Representations. New York: Columbia University Press.
 Conceptual Atomism and the Computational Theory of Mind

Chomsky, Noam. 1988. “Language and the problems of knowledge.” In The Philosophy of Lan-
guage, A.P. Martinich (ed), 509–527. New York: Oxford University Press.
Chomsky, Noam. 1998. On Language. New York: The New Press.
Coulter, Jeff. 1983. Rethinking Cognitive Theory. New York: St. Martin’s Press.
Crane, Tim. 1991. “Social Content and Psychological Content.” In The Twin-Earth Chronicles, A.
Pessin and S. Goldberg (eds), 284–304. Armonk, NY: M.E. Sharpe.
Churchland, Paul. 1981. “Eliminative materialism and the propositional attitudes.” In D.
Rosenthal (ed.), The Nature of Mind, 601–612. New York: Oxford University Press.
Cresswell, Max. “Structured meanings.” In Meaning and Truth, J. Garfield and M. Kiteley (eds),
446–452.
Davidson, Donald. 1967. “Truth and Meaning.” In The Philosophy of Language, A.P. Martinich
(ed), 91–101. Oxford University Press.
Davidson, Donald. 1980. Essays on Actions and Events. New York: Oxford University Press.
Davidson, Donald. 1980b. Inquiries into Truth and Interpretation. New York: Oxford University
Press.
Davidson, Donald. 1987. “Knowing One’s Own Mind.” In The Twin-Earth Chronicles, A. Pessin
and S. Goldberg (eds), 323–341. Armonk, NY: M.E. Sharpe.
Davidson, Donald. 2004. Problems of Rationality. New York: Oxford University Press.
Davidson, Donald. 2005. Subjective, Intersubjective, Objective. New York: Oxford University
Press.
Dennett, Daniel. Brainstorms. 1978. Cambridge: MIT press.
Donnellan, Keith. 1966. “Reference and Definite Descriptions.” In A.P. Martinich (ed). The Phi-
losophy of Language, 235–257. New York: Oxford University Press.
Donnellan, Keith. 1974. “Speaking of nothing.” The Philosophical Review 74: 3–31.
Dretske, Fred. 1982. Knowledge and the Flow of Information. Cambridge: MIT Press.
Dummett, Michael. 1978. Truth and Other Enigmas. Cambridge: Harvard University Press.
Dummett, Michael. 1973. Frege: Philosophy of Language. Cambridge: Harvard University Press.
Evans, Gareth. 1982. The Varieties of Reference. Oxford: Clarendon Press.
Evans, Gareth. 1985. Collected Papers. Oxford: Clarendon Press.
Falvey, Kevin. 1994. “Externalism, Self-Knowledge, and Skepticism.” Unpublished Dissertation.
Department of Philosophy. University of Minnesota.
Falvey, Kevin and Owens, Joseph. 1994. “Externalism, Self-Knowledge, and Skepticism.” Philo-
sophical Review 103: 107–137.
Field, Hartry. 1977. “Logic, Meaning, and Conceptual Role.” Journal of Philosophy 69: 378–408.
Fodor, Jerry. 1968. Psychological Explanation. New York: Random House.
Fodor, Jerry. 1975. The Language of Thought. New York: Thomas Y. Crowell
Fodor, Jerry. 1981. “Methodological Solipsism Considered as a Research Strategy in Cognitive
Psychology.” In D. Rosenthal (ed.), The Nature of Mind, 485–498. New York: Oxford Uni-
versity Press.
Fodor, Jerry. 1981b. Representations. Cambridge: The MIT Press.
Fodor, Jerry. 1983. The Modularity of Mind. Cambridge: The MIT Press.
Fodor, Jerry. 1987. Psychosemantics. Cambridge: The MIT Press.
Fodor, Jerry. 1987b. “Individualism and Supervenience.” In The Twin-Earth Chronicles, A. Pessin
and S. Goldberg (eds), 192–218. Armonk, NY: M.E. Sharpe.
Fodor, Jerry. 1990. A Theory of Content and Other Essays. Cambridge. The M.I.T. Press.
 Bibliography 

Fodor, Jerry. 1994. “Fodor.” In S. Guttenplan (ed.), A Companion to the Philosophy of Mind,
292–300. Oxford: Blackwell
Fodor, Jerry. 1994b. The Elm and the Expert: Mentalese and Its Semantics. Cambridge: M.I.T.
Press.
Fodor, Jerry. 1998. Concepts. Clarendon Press: Oxford.
Fodor, Jerry and Lepore, Ernest. 2002. The Compositionality Papers. New York: Oxford Univer-
sity Press.
Fodor, Jerry and Pylyshin, Zenon. 1988. “Connectionism and cognitive architecture: a critical
analysis.” In Connections and Symbols, S. Pinker (ed.), 3–72. Amsterdam: Elsevier.
Frege, Gottlob. 1884. Grundlagen der Arithmetik, eine logisch-mathematische Untersuchung über
den Begriff der Zahl. Breslau: W. Koebner.
Frege, Gottlob. 1891. “On Function and Concept.” In The Frege Reader, Michael Beaney (ed),
130–148. Oxford: Blackwell.
Frege, Gottlob. 1892. “On Sinn and Bedeutung.” In The Frege Reader, Michael Beaney (ed), 151–
171. Oxford: Blackwell.
Frege, Gottlob. 1892b. “On Concept and Object.” In The Frege Reader, Michael Beaney (ed),
181–193. Oxford: Blackwell.
Frege, Gottlob. 1918. “Negation.” In The Frege Reader, Michael Beaney (ed), 346–361. Oxford:
Blackwell.
Frege, Gottlob. 1918b. “Thought.” In The Frege Reader, Michael Beaney (ed), 325–345. Oxford:
Blackwell.
Frege, Gottlob. 1954. The Foundations of Arithmetic. Translated by J.L. Austin. New York: Oxford
University Press.
Freud, Sigmund. 1915. “The unconscious.” In P. Rieff (ed) General Psychological Theory, 116–
150. New York: Macmillan Publishing Co.
Freud, Sigmund. 1965. New Introductory Lectures on Psychoanalysis. London: W.W. Norton and Co.
Freud, Sigmund. 1998. The Interpretation of Dreams. New York: Basic Books.
Garfield, Jay. 1994. “Modularity.” In S. Guttenplan (ed) A Companion to the Philosophy of Mind,
441–448.
Gödel, Kurt. 1953. “Is mathematics syntax of language?” In Kurt Gödel: Collected Works, Volume
III, S. Feferman (ed.), 334–356. New York: Oxford.
Grice, H.P. 1957. “Meaning.” In The Philosophy of Language, A.P. Martinich (ed), 72–78. New
York: Oxford University Press.
Grush, Rick. 1999. Guide to Gareth Evans’ The Varieties of Reference. Unpublished e-manu-
script (http://mind.ucsd.edu/mix.resources/evans/chapterguides).
Hacker, P.M.S. 1999. Wittgenstein on Human Nature. New York: Routledge.
Hacker, P.M.S. 1996. Wittgenstein: Mind and Will. Oxford: Blackwell.
Hacker, P.M.S and Baker, Gordon. 1980. Wittgenstein: Understanding and Meaning. Oxford:
Blackwell.
Hacker, P.M.S and Baker, Gordon. 1984. Scepticism, Rules, and Language. Oxford: Blackwell.
Hacker, P.M.S and Baker, Gordon. 1984b. Language, Sense, and Nonsense. Oxford: Blackwell.
Hacker, P.M.S and Baker, Gordon. 1985. Wittgenstein: Rules, Grammar, and Necessity. Oxford:
Blackwell.
Hacking, Ian. 1975. Why does language matter to philosophy? Cambridge: Cambridge University
Press.
Hale, Bob and Wright, Crispin (eds). 1997. A Companion to the Philosophy of Language. Oxford:
Basil Blackwell.
 Conceptual Atomism and the Computational Theory of Mind

Hart, H.L.A. 1961. The Concept of Law. Oxford: The Clarendon Press.
Hempel, Carl Gustav. 1952. Fundamentals of Concept Formation in Empirical Science. Chicago:
University of Chicago Press.
Hempel, Carl Gustav. 1965. Aspects of Scientific Explanation. New York: The Free Press.
David Hershenov: 2002. A Defense of the Biological Approach to Personal Identity. Unpublished
dissertation. Department of Philosophy, University of California at Santa Barbara.
David Hershenov: 2005. “Do Dead Bodies Pose a Problem for Biological Approaches to Per-
sonal Identity?” In Mind (114): 31–59
Horst, Stephen. 1996. Symbols, Computation, and Intentionality. Berkeley: University of Califor-
nia Press.
Jackendoff, Ray. 1989. “What is a concept, that people may grasp it?” In Concepts: Core Readings,
E. Laurence and S. Laurence (eds), 305–334. Cambridge: MIT Press.
Jackson, Frank and Pettit, Philip. 2004. “Causation in the Philosophy of Mind”. In Mind, Moral-
ity, and Explanation, F. Jackson, P. Pettit, and M. Smith (eds), 45–68. Oxford: Oxford Uni-
versity Press.
Jackson, Frank and Pettit, Philip. 2004b. “Some content is narrow.” In Mind, Morality, and Expla-
nation, F. Jackson, P. Pettit, and M. Smith (eds), 69–94. Oxford: Oxford University Press.
Jackson, Frank and Pettit, Philip. 2004c. “Functionalism and Broad Content” In Mind, Morality, and
Explanation, F. Jackson, P. Pettit, and M. Smith (eds), 95–118. Oxford: Oxford University Press.
Jackson, Frank and Pettit, Philip. 2004d. “Program Explanation: A General Perspective”. In
Mind, Morality, and Explanation, F. Jackson, P. Pettit, and M. Smith (eds), 119–130. Oxford:
Oxford University Press.
Kaplan, David. “Quantifying In.” 1968. In The Philosophy of Language, A.P. Martinich (ed), 370–
391. New York: Oxford University Press.
Kaplan, David, 1975. “Dthat”. In The Philosophy of Language, A.P. Martinich (ed), 316–329. New
York: Oxford University Press.
Kaplan, David. 1975b. “How to Russell a Frege-Church.” Journal of Philosophy 72: 716–729.
Kaplan, David. 1978. “On the Logic of Demonstratives.” Journal of Philosophical Logic 8: 81–98.
Kaplan, David. 1989. “Demonstratives.” In Themes from Kaplan, J. Almog, J. Perry,, H. Wettstein
(eds), 481–564. New York: Oxford University Press.
Kaplan, David. 1989b. “Afterthoughts.” In Themes from Kaplan, J. Almog, J. Perry,, H. Wettstein
(eds), 565–614. New York: Oxford University Press.
Katz, Jerrold. 1972. Semantic Theory. New York: Harper and Row.
Kim, Jaegwon. 1993. Supervenience and Mind. Cambridge: Cambridge University Press.
King, Jeffrey C., 1995, “Structured Propositions and Complex Predicates”, Nous 29(4), 516–535
King, Jeffrey C., 1996, “Structured Propositions and Sentence Structure”, Journal of
Philosophical Logic 25: 495–521.
King, Jeffrey C., 1997, “Structured Propositions” in the Stanford Encyclopedia of Philosophy. On-
line publication. http://plato.stanford.edu/entries/propositions-structured/
Kripke, Saul. 1971. “Identity and Necessity.” In Identity and Individuation, M. K. Munitz (ed).
New York: New York University Press.
Kripke, Saul. 1972. Naming and Necessity. Cambridge: Harvard University Press.
Kripke, Saul. 1977. “Speaker’s Reference and Semantic Reference.” In A.P. Martinich (ed). The
Philosophy of Language, 248–266. New York: Oxford University Press.
Kripke, Saul. 1979. “A Puzzle about Belief.” In Meaning and Use, A. Margalit (ed). Dordrecht:
Reidel.
 Bibliography 

Kripke, Saul. 1982. Wittgenstein on Rules and Private Language. Cambridge: Harvard University
Press.
Kuczynski, John-Michael. 2007. “Intensionality”, forthcoming in the Journal of Pragmatics.
Kuczynski, John-Michael. 2006. “the Analogue-Digital Distinction and the Cogency of Kant’s
Transcendental Arguments.” Existentia Meletai Sophias 16: 279-320.
Kuczynski, John-Michael. 2007b. “Does Possible World Semantics Turn all Propositions into
Necessary Ones.”? Journal of Pragmatics 39: 872-916.
Kuczynski, John-Michael. 2007c. “Did Berkeley Refute the Myth of the Given Before Sellars?”
Unpublished manuscript.
Kuczynski, John-Michael. 2006b. “Two Concepts of Sentential ‘Form’ and the So-called Compu-
tational Theory of Mind.” Philosophical Psychology 19 (6): 1–27.
Kuczynski, John-Michael. 2006c. “Formal Operations and Simulated Thought.” Philosophical
Explorations 9 (2): 221–234.
Kuczynski, John-Michael. 2006d. “On Marga Reimer’s Descriptions and Beyond.” Pragmatics and
Cognition 14 (1): 196–204.
Kuczynski, John-Michael. 2005. “The Concept of a Symbol and the Vacuousness of the Sym-
bolic Conception of Thought.” Semiotica 154 (4): 243–264.
Kuczynski, John-Michael. 2005b. “Must one Know a Language to Grasp Propositions?” Teorema
24 (2): 43–65.
Kuczynski, John-Michael. 2005c. Literal Meaning and Cognitive Content. Unpublished manuscript.
Kuczynski, John-Michael. 2004. “A non-Russellian Treatment of the Referential-Attributive Dis-
tinction.” Pragmatics and Cognition. 12 (2): 253–294
Kuczynski, John-Michael. 2004b. “Why Definite Descriptions Really are Referring Expressions.”
Grazer Philosophische Studien 68: 45–79.
Kuczynski, John-Michael. 2004c. “Another Argument against the Thesis that there is a Language
of Thought.” Communication and Cognition 37 (2): 83–104.
Kuczynski, John-Michael. 2004d. “Non-declarative Sentences and the Theory of Descriptions.”
Principia (8) 1: 119–154.
Kuczynski, John-Michael. 2003. “Some Arguments against Intentionalism.” Acta Analytica 19
(32): 107–142.
Kuczynski, John-Michael. 2002. “Does the Idea of a ‘Language of Thought’ Make Sense?” Com-
munication and Cognition 35 (4): 173–192.
Langford, C.H. 1942. “The notion of analysis in Moore’s philosophy.” In P.A. Schilpp (ed) The
Philosophy of G.E. Moore, 321–342. LaSalle: Open Court.
Lepore, Ernie. 1994 “Conceptual Role Semantics.” In A Companion to the Philosophy of Mind, S.
Guttenplan (ed), 193–200, 193–199. Blackwell: Cambridge, MA.
Lewicki, Pawel and Hill, Thomas and Czyzewska, Maria. 1992. “Nonconscious acquisition of
information.” American Psychologist 47 (6): 796–801.
Lewis, Clarence Irving. 1946. An Analysis of Knowledge and Valuation. Cambridge: Harvard
University Press.
Lewis, Clarence Irving. 1952. “The modes of meaning.” In Semantics and the Philosophy of Lan-
guage. L. Linsky (ed), 50–66. Chicago: University of Illinois Press.
Lewis, David. 1972 “Psychophysical and Theoretical Identifications.” In D. Rosenthal (ed.), The
Nature of Mind, 204–210. New York: Oxford University Press.
Lewis, David. 1980. “Mad pain and Martian pain.” In D. Rosenthal (ed.), The Nature of Mind,
229–234. New York: Oxford University Press.
 Conceptual Atomism and the Computational Theory of Mind

Lewis, David. 1975. “Language and Languages.” In The Philosophy of Language, A.P. Martinich
(ed), 489–508. Oxford: Oxford University Press.
Lewis, David. 1984. On the Plurality of Worlds. Oxford: Blackwell.
Loar, Brian. 1985. “Social Content and Psychological Content.” In The Twin-Earth Chronicles, A.
Pessin and S. Goldberg (eds), 180–191. Armonk, NY: M.E. Sharpe.
Loewer, Barry. 1999. ‘A Guide to Naturalizing Semantics.” In A Companion to the Philosophy of
Language, B. Hale and C. Wright (eds), 108–126. Oxford: Blackwell.
Lowe, E.J. 1996. Subjects of Experience. Cambridge: Cambridge University Press.
Lycan, William. 1984. Logical Form in Natural Language. Cambridge: MIT Press.
Lyons, John. 1977. Semantics. Cambridge: Cambridge University Press.
Mates, Benson. 1952. “Synonymy.” In Semantics and the Philosophy of Language. L. Linsky (ed.),
111–138. Chicago: University of Illinois Press.
McCawley, James D. 1981. Everything that Linguists Have Always Wanted to Know about Logic.
Chicago: University of Chicago Press.
McCawley, James D. 1998. The Syntactic Phenomena of English. Chicago: University of Chicago
Press.
McCulloch, Gregory. 1989. The Game of the Name. Clarendon: Oxford.
McDowell, John. 1994. Mind and World. Cambridge: Harvard University Press.
McDowell, John. 1998. Meaning, Knowledge & Reality. New York: Oxford University Press.
McGinn, Colin. 1988. Mental Content. New York: Oxford University Press.
McGinn, Colin. 2002. The Making of a Philosopher. New York: Harper Collins.
McKinsey, Michael. 1991. “Anti-individualism and privileged access.” In The Twin-Earth Chron-
icles, A. Pessin and S. Goldberg (eds), 355–361. Armonk, NY: M.E. Sharpe.
Merricks, Trenton. 2001. Objects and Persons. New York: Oxford University Press.
Montague, Richard. 1974. “The Proper Treatment of Quantification in Ordinary English.” In
Formal Philosophy, Richard Thomason (ed). New Haven: Yale University Press.
Moore, Robert. C. 1995. Logic and Representation. Palo Alto: CSLI Publications.
Nagel, Ernest. 1962. The Structure of Science. Indianapolis: Hackett
Neale, Stephen. 1990. Descriptions. Cambridge: The M.I.T. Press.
Neri-Castañeda, Hector. 1989. Thinking, Language, and Experience. Minneapolis: University of
Minnesota Press.
Pap, Arthur. 1958. Semantics and Necessary Truth. New Haven: Yale University Press.
Pap, Arthur. 1962. Introduction to the Philosophy of Science. New Haven: Yale University Press.
Peacocke, Christopher. 1989. “Perceptual Content.” In Themes from Kaplan, J. Almog, J. Perry,,
H. Wettstein (eds), 297–330. Oxford: Oxford University Press.
Peacocke, Christopher. 1992. A Study of Concepts. Cambridge: MIT Press.
Peacocke, Christopher. 1996. “Précis of A Study of Concepts.” In Concepts, E. Margolis and S.
Laurence (eds), 335–338.
Perry, John. 2000. The Problem of the Essential Indexical and Other Essays. Palo Alto: CSLI Pub-
lications.
Plantinga, Alvin. 1974. The Nature of Necessity. New York: Oxford University Press.
Platts, Marc. 1979. The Ways of Meaning. London: Routledge & Kegan Paul.
Putnam, Hilary. 1975. “The Meaning of ‘Meaning’”. In The Twin-Earth Chronicles, A. Pessin and
S. Goldberg (eds), 1–52. Armonk, NY: M.E. Sharpe.
Putnam, Hilary. 1996. “Introduction to The Twin-Earth Chronicles.” In The Twin-Earth Chroni-
cles, A. Pessin and S. Goldberg (eds), xv-xxii. Armonk, NY: M.E. Sharpe.
 Bibliography 

Pylyshin, Zenon W. 1984. Computation and Cognition. Cambridge: The M.I.T. Press.
Quine, W.V.O. 1951. “Two dogmas of empiricism.” In The Philosophy of Language, A.P. Martin-
ich (ed), 26–39. Oxford: Oxford University Press.
Quine, W.V.O. 1953. From a Logical Point of View. Cambridge: Harvard University Press.
Quine, W.V.O. 1960. Word and Object. Cambridge: MIT Press.
Quine, W.V.O. 1970. Philosophy of Logic. Cambridge: Harvard University Press.
Recanati, François. 2003. Literal Meaning. Cambridge: Cambridge University Press.
Rescorla, Michael. 2003. “Is Thought Explanatorily prior to Language?” Unpublished Disserta-
tion. Department of Philosophy. Harvard University.
Rey, George. 1996. “Resisting primitive compulsions.” In Concepts, E. Margolis and S. Laurence
(eds), 339–343..
Richard, Mark. 1991. Propositional Attitudes. Cambridge: Cambridge University Press.
Rosenthal, David M. 1986. “Two concepts of consciousness.” In D. Rosenthal (ed), The Nature of
Mind, 462–477. New York: Oxford University Press.
Russell, Bertrand. 1903. Principles of Mathematics. Cambridge: Cambridge University Press.
Russell, Bertrand. 1905. “On Denoting.” In The Philosophy of Language, A.P. Martinich (ed),
203–211. The Philosophy of Language Oxford: Oxford University Press.
Russell, Bertrand. 1917. Mysticism and Logic. London: George Allen Unwin.
Bertrand, Russell. 1918. “The philosophy of logical atomism.” In Logic and Knowledge, R.C.
Marsh (ed), 175–282.
Russell, Bertrand. 1920. Introduction to Mathematical Philosophy. London: George Allen Unwin.
Russell, Bertrand. 1921. The Analysis of Mind. London: George Allen Unwin
Russell, Bertrand. 1927. An Outline of Philosophy. London: George Allen Unwin
Russell, Bertrand. 1957. Mysticism and Logic. Garden City, New Jersay: Doubleday.
Russell, Bertrand. 1948. Human Knowledge: Its Scope and Limits. London: George Allen Unwin.
Russell, Bertrand. 1950. “Logical positivism.” In Logic and Knowledge, R.C. Marsh (ed), 367–382.
Russell, Bertrand. 1959. My Philosophical Development. London: George Allen Unwin.
Russell, Bertrand. 1984. Theory of Knowledge. London: George Allen Unwin.
Salmon, Nathan. 1986. Frege’s Puzzle. Cambridge: M.I.T. Press.
Salmon, Nathan. 1989. “Tense and Singular Propositions.” In Themes from Kaplan, J. Almog, J.
Perry,, H. Wettstein (eds), 331–332. New York: Oxford University Press.
Salmon, Nathan. 2005. “Two conceptions of semantics.” In Semantics and Pragmatics, Z. Szabo
(ed), 317–328. Ithaca: Cornell University Press.
Salmon, Nathan. 2005. Metaphysics, Mathematics, and Meaning. New York: Oxford. University
Press.
Salmon, Wesley. 1984. Scientific explanation and the causal structure of the world. Princeton:
Princeton University Press.
Sartre, Jean-Paul. 1943. Being and Nothingness. New York: McMillan and co.
Saussure, Ferdinand de. 1966. Course in General Linguistics. New York: McGraw-Hill Book
Company.
Searle, John. 1969. Speech Acts. Cambridge: Cambridge University Press.
Searle, John. 1979. Expression and Meaning. Cambridge: Cambridge University Press.
Searle, John. 1983. Intentionality: an Essay on the Philosophy of Mind. Cambridge: Cambridge
University Press.
Searle, John. 1984. Minds, Brains, and Science. Cambridge: Harvard University Press.
Searle, John. 1992. The Rediscovery of the Mind. Cambridge: MIT Press.
Sellars, Wilfred. 1963. Science, Perception and Reality. New York: Routledge & Kegan Paul.
 Conceptual Atomism and the Computational Theory of Mind

Shoemaker, Sydney. “Parfit on Identity.” In Jonathan Dancy (ed.), Reading Parfit. Oxford: Basil
Blackwell.
Shope, R.K. 1983. The Analysis of Knowing. Princeton: Princeton University Press.
Sklar, Lawrence. 1974. Space, Time, and Spacetime. Berkeley: University of California Press.
Soames, Scott. 1989. “Direct Reference and Propositional Attitudes.” In Themes from Kaplan, J.
Almog, J. Perry, H. Wettstein (eds), 393–420. New York: Oxford University Press.
Soames, Scott. 2001. Beyond Rigidity. Princeton: Princeton University Press.
Soames, Scott. 2005. Reference and Description: The Case Against Two-dimensionalism. Prince-
ton: Princeton University Press.
Stich, Stephen P. 1978. “Scientific versus folk psychology.” In D. Rosenthal (ed), The Nature of
Mind, 591–600. New York: Oxford University Press.
Stich, Stephen P. 1983. From Folk Psychology to Cognitive Science: A Case Against Belief. Cam-
bridge: the MIT Press.
Strawson, Galen. 1994. Mental Reality. Cambridge: MIT Press.
Strawson, Peter. 1950. “On Referring.” In The Philosophy of Language, A.P. Martinich (ed.), 219–
234. Oxford: Oxford University Press.
Strawson, Peter. 1959. Individuals. New York: Routledge.
Strawson, Peter. 1969. “Meaning and Truth.” In The Philosophy of Language, A.P. Martinich (ed.),
91–101. Oxford: Oxford University Press.
Stroud, Barry. 2001. The Quest for Reality. New York: Oxford University Press.
Taylor, Kenneth. 1998. Truth & Meaning. Oxford: Blackwell.
Tughendat. Ernst. 1970. “The Meaning of ‘Bedeutung’ in Frege.” In Analysis 30: 177–189.
Wittgenstein, Ludwig. 1922. Tractatus Logico-philosophicus. London: Routledge, Kegan & Paul.
Wittgenstein, Ludwig. 1958. The Philosophical Investigations. Oxford: Blackwell.
Wittgenstein, Ludwig. 1974. Philosophical Grammar. Oxford: Blackwell.
Wittgenstein, Ludwig. 1980. Remarks on the Foundations of Psychology. Volume 1. Chicago: Uni-
versity of Chicago Press.
Wittgenstein, Ludwig. 1980b. Culture and Value. Oxford: Blackwell.
Wittgenstein, Ludwig. 1983. Remarks on the Foundations of Mathematics. Cambridge: M.I.T.
Press.
 Index
A semantic externalism, 4
Algorithms.
See Mechanical procedures
Analogue-digital distinction:
419–460
Analytic Functionalism: 255–259
See also Lewis, David
Analytic truth:
in relation to Kripke’s work,
97–103
in relation to formal truth,
238–241
Aristotle: 193, 286, 494
Armstrong, David: 430–431
Articulation-problem: 432
See also Hacking, Ian
Ayer, A.J.: 360, 363
B 498–502
Bach, Kent:
on the Theory of Descrip-
tions, 74
on literal meaning, 271
Baker, Gordon: 326, 359
See also Hacker, P.M.S.
Barwise, Jon:
on indexicals and logical
form, 15–16, 237
on the predicative nature of
sense-perception, 25
on the nature of entailment,
477
on situation-semantics, 477
See also Perry, John
Berkeley, George:
on conceptual non-atomism,
10, 217
on the relationship between
phenomenology and rep-
resentational content, 32
on the determinate nature of
perceptual content, 317
Blackburn, Simon:
on the difference between
content-externalism and
on the relation between
cognitive and linguistic
content, 37
on the Theory of Descrip-
tions, 82
on functionalism, 212
on the relation between what
can be known and what
can be said, 322
on Grice’s semantics, 417,
483, 490
as advocate of a quasi-
Gricean semantics,
498–502
his semantics in relation to
the compositional charac-
ter of linguistic meaning,
Boguslawski, Andrzej: 108
Bonjour, Laurence:
on conceptual non-atomism,
10, 217
on Quine on analyticity, 307
Brandom, Robert:
on inferential role semantics,
224, 331, 427
on Frege, 331–332
on meaning-holism, 421
on the structures of proposi-
tions, 441
Burge, Tyler:
on content-externalism, 4, 9,
29, 409
on narrow content, 9
on the individuation of men-
tal entities, 34
on conception, 54
on self-knowledge, 164–167
on why conception cannot be
understood in strictly de-
scriptivist terms, 184–186
See also Conception.
C
Carnap, Rudolph:
on logical truth, 14
on the non-atomistic con-
ception of one’s grasp of
theoretical entities, 220
on partial explanations, 296
on the reducibility of all
sciences to microphysics,
344, 380
on induction, 406
on the structures of proposi-
tions, 447
Chalmers, David: 9
Chomsky, Noam:
on the existence of subper-
sonal thought, 31, 323–326
on behaviorism, 254, 486
on Quine on analytic truth,
305
on the distinction between
performance and compe-
tence, 336, 399
debate with Searle concern-
ing subpersonal thought,
343–358
on the falsity of dispositional-
ist analyses of proposi-
tional attitudes, 376
on depth-grammar, 420–421
on the falsity of behaviorist
analyses of propositional
attitudes, 486
on Grice, 498
Churchland, Paul:
on self-knowledge, 170, 338
on folk-psychology, 317
on subpersonal thought, 323
Cognitive maps: 42, 109–121
See also Evans, Gareth.
Computational Theory of Mind:
definition of, 10, 222
arguments for, 222–223
in relation to the causal po-
tency of representational
 Conceptual Atomism and the Computational Theory of Mind
content, 7–8, 189–216
in relation to indexicality,
232–236
as involving an equivocal
use of expressions such as
“form”, “algorithm”, and
“mechanical procedure”,
232–236
Conception:
why it must be understood in
descriptivist terms, 41–108
why the causal theory is con-
sistent with the descriptiv-
ist theory of conception,
109–123
why a descriptivist theory of
conception is compatible
with a non-descriptivist,
Kripke-style semantics,
109–123
in relation to the problem
of forgotten descriptive
information, 123–129
in relation to Frege’s paradox,
125–126
as distinguished from recog-
nition, 126–127
why the causal theory of con-
ception involves a failure
to distinguish the mere
permissibility of inferences
from the actual making of
them, 230–231
Concepts.
See Conception.
See also Non-conceptual
content.
Conceptual analysis.
See Paradox of analysis.
Conceptual Atomism:
definition of, 10–11
in relation to content-exter-
nalism, 10–11
in relation to Fodor’s work,
10–11
in relation to the Compu-
tational Theory of Mind,
12–13, 216–260
Conceptual Role Semantics.
See Inferential Role Seman-
tics.
Connectionism:
in relation to mental causa-
tion and personal identity,
343
in relation to the viability of
the Computational Theory
of Mind, 399
in relation to the systematic-
ity and productivity of
thought, 399
Content:
as distinguished from truth-
maker, 19–20
as distinguished from post-
perceptual meta-data, 20
See also Content-exter-
nalism.
Content-externalism:
in relation to self-knowledge,
163–180, 507
in relation to representational
content, 3–216
in relation to psychological
explanation, 5, 137–141
Coulter, Jeff: 249
Cresswell, Max: 444, 451–457
D on de re senses, 125–148
Data.
See Post-perceptual meta-
data.
Davidson, Donald:
on Frege’s semantics, 93
on self-knowledge, 164,
170–177
on Frege on propositions, 446
on the need to posit non-
conceptual content, 475
on why linguistic knowledge
is constitutive of cognitive
ability, 485–486
Declarative knowledge: 325–326
See also Procedural knowl-
edge.
Dennett, Daniel: 228, 360, 409
Depth-grammar: 419–420
Designation. See Reference.
Determinates versus determina-
bles: 430–431
De re senses: 129–152
why the concept of a de re
sense is a conflation of
epistemology and seman-
tics, 141–150
in relation to Kripke’s Padare-
wski paradox, 146–147
Searle on, 147–148
McDowell on Searle on,
150–151
See also Evans, Gareth.
Discourse-theory: 412
Disjunctivism: 36–37
See also McDowell, John.
Donnellan, Keith: 49, 109
Dretske, Fred: 116–117
Dummett, Michael:
on wide-scope descriptiv-
ism, 53
on Quine on analyticity, 307
on reference and contextual
definition, 491
E
Evans, Gareth:
on narrow content, 9
on conception and its relation
to literal meaning, 20
on the existential character of
perceptual information, 36
on semantic knowledge, 44
on “Julius”, 49–51
on cognitive maps, 129–152
on the photograph-model of
conception, 129–130
on non-conceptual content,
316, 423–426, 443, 507
on subpersonal thought,
335–338
on the non-frozen character
of concepts, 350
Extensionality.
See Substitution-failures.
F
Falvey, Kevin: 167–170
Field, Hartry: 224, 427
Fodor, Jerry:
on narrow content, 9
on the causal theory of con-
ception, 9
on the Computational Theory
of Mind, 9, 217–261
on the Symbolic Conception
of Thought
on syntax, 13–17
on McDowell on mental
causation, 22
on the modularity of mind,
33, 351

on Inferential Role Seman-
tics, 223–225, 428
on behaviorism, 254
his arguments for conceptual
atomism, 293–333
his arguments for the
Symbolic Conception of
Thought, 297–408
on conceptual analysis,
319–320
on subpersonal thought, 326
on meaning-holism, 428
on our intuitions concerning
analyticity, 468
on our concepts of abstract
entities, 476
on why non-social creatures
can think, 485
Formal truth: 13–15, 235, 237–241
See also Analytic truth.
Fraassen, Bas van: 220
Frege, Gottlob:
on compositionality, 84;
on substitution-failures,
84–95
on identity-claims, 98
failure to distinguish between
semantics and pre-seman-
tics, 104–105
failure to distinguish between
pre-semantics and pre-pre-
semantics, 105–107
on the public nature of
senses, 153
on cognitive dynamics, 330
as alleged advocate of In-
ferential Role Semantics,
331–332
on the structures of proposi-
tions, 446–447
on the difference between
logical and grammatical
form, 449
on the unity of the proposi-
tion, 449, 459–460
on the differences between
positive and negative at-
tributes, 464
on the differences between
singular and general state-
ments, 480
on the mind-independent
nature of truth, 486
on reference and contextual
definition, 491, 492
on number, 492
Freud, Sigmund:
on the differences between
ideographic and discursive
mentation, 129–130, 426
Wittgenstein on, 373
similarities between his
views and contemporary
“higher-order thought”
theories of consciousness,
377–378
on the unconscious, 382–384,
390–396
on repression, 390–396
on forgetting, 393
See also Sartre, Jean-Paul.
Functionalism:
as incompatible with the
causal potency of the
mental, 189–191
Jackson-Pettit defense of, 209
why it cannot distinguish be-
tween actual and simulated
thought, 211–213
Kripkenstein problems with,
211–213
Blackburn on, 212
Peacocke as advocate of,
471–472
See also Analytic Function-
alism.
G Hoffman, Dustin: 351
Gödel, Kurt: 14
Goodman, Nelson: 404
Grice, H.P.:
on knowledge of intentionally
affirmed semantic content,
155–156
Searle on, 151
fallacies in his semantics,
155–157, 416–418, 483–502
on the meaning of “mean-
ing”, 230
Blackburn on, 417, 483
his semantics in relation to
the compositional charac-
ter of linguistic meaning,
498–501
Grush, Rick: 36, 118, 129
Index 
H
Hacker, P.M.S.:
on attitudinal eliminativ-
ism, 213
on subpersonal thought, 326
on conceptual non-atomism,
351
on Wittgenstein on why con-
cepts are not frozen, 351
on Wittgenstein on private
languages and rule-follow-
ing, 359
See also Baker, Gordon.
Hacking, Ian: 432
Hart, H.L.A.: 242
Hempel, Carl Gustav:
on measurement, 177
on the irreducibility of theo-
retical to observational
propositions, 214
on the non-atomistic con-
ception of one’s grasp of
theoretical entities, 220
on instrumentalism and
operationalism, 220, 311
on attempts to use the
Ramsey-sentence to define
theoretical terms, 259
on analytic truth, 308–313
Hershenov, David: 339
Hilbert, David:
on uninterpreted formal
calculi, 472
on number, 492
Horst, Stephen:
on conception, 54
on symbolism, 231
on mechanical procedures,
249
Hume, David:
on personal identity, 339
on causality, 381, 482
as advocate of the view that
sense-perception has a
digital structure, 421
on memory, 437
on induction, 442
I
Indexicals:
in relation to pre-semantic
implicature, 18–19, 69–71
in relation to the distinction
 Conceptual Atomism and the Computational Theory of Mind
between tokens and types,
69–70
Induction: 404–407
Inferential Role Semantics:
223–225, 427–428
Intensionalism: 32–33, 38
See also Sellars, Wilfred. See
also Berkeley, George.
J on object-individuation, 65,
Jackendoff, Ray: 84
Jackson, Frank:
on narrow content, 8
on the causal potency of the
mental, 18
on the difference between
program-causality and ef-
ficient causality, 18, 189–213
on functionalism, 209
on prediction and explana-
tion, 379
See also Pettit, Philip.
K non-atomism, 10
Kaplan, David:
on semantic externalism, 4
on content-externalism, 4, 29
on direct-reference, 18, 107
on what one can mean, 49
on the causal theory of con-
ception, 49, 109, 116
on cognitive dynamics, 55,
330
on molecular sentences in
relation to the Theory of
Descriptions, 95
on Frege’s semantics, 105–106
on “Newman 1”, 109–110
on descriptivist theories of
perception in relation to
reduplication-problems,
149
on the relation between logi-
cal and syntactical form,
236–237
on formal truth, 238
on the nature of proposi-
tions, 419
Katz, Jerrold: 244
Kim, Jaegwon: 17, 23
King, Jeffrey C.: 444
King, Stephen: 263
Kripke, Saul:
on the semantics of proper
names, 41–42, 296, 469,
507–508
on the causal theory of con-
ception, 42–43
on the causal theory of refer-
ence, 42–43
on the circularity inherent in
Russell’s theory of proper
names, 47
298–299, 441
on ambiguity, 97
on a posteriori necessity,
97–99, 507
on inferential role semantics,
224
on Wittgenstein on private-
language and rule-follow-
ing, 359
on analyticity, 469–470
Kuczynski, John-Michael:
on Berkeley on conceptual
on Berkeley on the relation-
ship between phenomenol-
ogy and representational
content, 32
on intensionalism, 32, 172, 385
on the relation between pre-
semantic implicature and
the viability of the Theory
of Descriptions, 77
on anaphora, 79
on neo-Russellian misun-
derstandings concerning
pragmatics, 82
on the difference between
conceptual and semantic
analysis, 83
on compositionality, 84
on the relationship between
grammar and semantics,
84
on Davidson on Frege’s
semantics, 94
on the viability of Russell’s
semantics in relation to
non-indicative sentences,
95
on Neale, 95
on legal obligation, 196
on epistemological founda-
tionalism, 217–218
on the regress argument
against the Symbolic Con-
ception of Thought, 228,
280–281, 409
on symbolism, 231–232
on the relationship between
logical and grammatical
form, 236
on Wittgenstein on non-
atomism, 306
on laws as narrow-scope as-
surances of protections of
moral rights, 305
on Wittgenstein on private
languages, 365
on the Symbolic Conception
of Thought, 405
on McDowell on non-con-
ceptual content, 425
on causation, 442
on induction, 442
on Montague on contextual
definition, 448
on Possible Worlds Seman-
tics, 451–452
on molecular propositions,
465
on why there is non-social,
non-linguistic thought, 486
on the semantics of non-
assertoric sentences, 504
L
Langford, C.H.: 319, 478
Leibniz, Gottfried: 305
Lepore, Ernie:
on functionalism, 212
on inferential role semantics,
224–225, 428
on meaning-holism, 428
Lewicki, Pawel: 325
See also Procedural Knowl-
edge.
Lewis, Clarence Irving: 447
Lewis, David:
on analytic functionalism,
255–259
on language, 410–416, 490
on Possible Worlds Seman-
tics, 451–452
on propositions as properties
of worlds, 453–454

Linguistic communication:
as involving isomorphism,
not identity, of thought,
158–162
Grice on, 155–156, 409,
416–418, 483–502
general nature of, 483–506
Linguistic meaning:
Grice on, 155–156, 409,
416–418, 483–502
function-theoretic account
of, 410–416, 489–490
David Lewis on, 410–416, 490
Blackburn on, 417
nature of, 483–506
definition of, 493–495,
505–506
compositionality of, 495–502
why Grice’s theory cannot ac-
commodate the composi-
tionality of, 498–502
See also Linguistic symbol.
Linguistic symbol: 261–284
definition of, 273–274,
505–506
Loar, Brian: 8
Lowe, E.J.: 339
Lycan, William: 228, 409
M Horst on, 249
Marr, David: 362, 368, 379
Mates, Benson: 80
McCawley, James:
on quantifiers, 853
on syntax, 244
McDowell, John:
as advocate of hardline
content-externalism, 8
on the causal potency of
mental content, 9, 17, 21–22
on the non-existence of
subpersonal cognitive
processes, 31
as advocate of disjunctiv-
ism, 37
on the non-existence of
subpersonal cognitive
processes, 31
on views that liken Frege’s
semantic theories to Rus-
sell’s, 89
on the impossibility of
meaningful psychological
statements, 136
on de re senses, 141
debate with Searle concern-
ing conception, 147–151
on non-conceptual content,
423–426, 475
on why embeddedness in a
society is constitutive of
cognitive ability, 485
See also Evans, Gareth.
McGinn, Colin:
on mental content and
content-externalism, 4–6,
29
on exegetical issues relating
to Wittgenstein’s views on
private-language and rule-
following, 359
See also Mental Causality.
Meaning-holism: 427–428
See also Inferential Role
Semantics.
Meaning.
See Linguistic meaning.
Mechanical procedures:
definition of, 243, 245
in relation to the Compu-
tational Theory of Mind,
248–250, 507
Coulter on, 249
Wittgenstein on, 249
Meinong, Alexius: 38
Mental Causality.
See Content-externalism.
Merrick, Trenton: 64, 339, 435
Montague, Richard:
on contextual definition,
448–449
on whether non-lexical items
refer, 492
Moore, Robert C.: 419, 444
N Perry, John:
Nagel, Ernest: 220
Neale, Stephen:
on the Theory of
Descriptions, 78–79, 82, 95
on quantification, 85–86
Nietzsche, Frederick: 393
Non-conceptual content:
419–466
in relation to the Symbolic
Conception of Thought,
26, 419–466
Index 
in relation to Fodor’s argu-
ment for conceptual atom-
ism, 316–322
Evans on, 316, 423–426
McDowell on, 317, 423–426
debate between Evans and
McDowell concerning,
423–426
O
Owens, Joseph: 167–170
See also Falvey, Kevin.
P
Padarewski paradox: 146–147,
174
See Psychological explana-
tion.
See also Kripke, Saul.
Pap, Arthur:
on measurement, 177
on instrumentalism and
operationalism, 220
on behaviorism and inter-
vening-variable theories in
psychology, 258, 375
on Leibniz on necessary
truth, 305
on induction, 406
Paradox of Analysis: 319–320,
478–479
See also Non-conceptual
content.
Peacocke, Christopher:
on conception, 296, 319,
467–481
on inferential role semantics,
331–333
Pears, David: 130
Perception.
See Sense-perception.
Personal identity: 339–342
on indexicals and logical
form, 15–16, 237
on the nature of entailment,
477
situation-semantics, 477
on the structures of proposi-
tions, 444
See also Barwise, Jon.
Pettit, Philip:
on narrow content, 8
on the causal potency of the
 Conceptual Atomism and the Computational Theory of Mind
mental, 18
on the difference between
program-causality and ef-
ficient causality, 18, 189–213
on functionalism, 209
on prediction and explana-
tion, 370
See also Jackson, Frank.
Picture-theory of linguistic
meaning: 426, 445
See also Wittgenstein,
Ludwig.
Platts, Mark: 498
Platonism: 412
Poincaré, Henri: 7
Post-perceptual meta-data:
20–21, 507
Pre-semantic implicature: 18–19,
26–28
Procedural knowledge: 325–326
See also Declarative knowl-
edge.
Productivity of language and
thought: 400–401
See also Systematicity of
language and thought.
Propositions: 429–466
as unordered sets, 443–444
as ordered sets, 444–445
Frege’s views on, 445–446
as sets of their own conse-
quences, 447–448, 450–451
Kuczynski’s definition of,
454–456
the unity of the proposition,
456–459
molecular propositions,
457–458, 463–465
See also Truth.
Psychological explanation:
in relation to content-exter-
nalism, 137–140
in relation to the Symbolic
Conception of thought,
275–284
Fodor’s syntactic analysis of,
278–280
Putnam, Hilary:
on program-causality, 192
on semantic externalism, 409
as late convert to content-
externalism, 409
Pylyshin: 271, 366, 399
Q between regulatory and
Quine, Willard van Orman:
on the modal status of
identity-claims, 99
on behaviorism and attitudi-
nal eliminativism, 213
on formal truth, 238
on analytic truth, 294,
301–308
R on substitution-failures,
Ramsey, Frank: 255–259
Recognition: 56–57
Reichenbach, Hans:
on dynamic versus kinematic
relativity, 7
on the relation between
spatiotemporal order and
causal connectedness,
113, 342
Reference:
in relation to quantification,
82–92
direct, 82–84, 87
indirect, 83–84, 87
definition of, 83, 88, 491, 502
Russell on, 89–90
Frege on, 91–92
rigid versus non-rigid, 91–92
Rescorla, Michael: 130, 426
Rosenthal, David: 377
Rey, George:
on Peacocke on conception,
296, 468–471
on Quine on analyticity, 469
Rules:
Searle on why an understand-
ing of rules demands a
rejection of Grice’s seman-
tics, 155–157, 483–484, 495
Searle on why an understand-
ing of rules demands
a rejection of the view
that there is subpersonal
thought, 326, 335, 338–358
Wittgenstein on rules,
359–373
Wittgenstein on why there
can be no linguistic rules
in non-social contexts,
359–365
Wittgenstein on the following
of rules, 367–380
Searle on the difference
constitutive rules, 488
See also Semantic rules.
Russell, Bertrand:
on the difference between
content and truth-maker,
19, 36, 436
on Meinong, 38
on object-involving proposi-
tions, 58–68
73–84
on the Theory of Descrip-
tions, 73–84
on identity-claims, 98, 102
on linguistic communication,
154, 159
on epistemological founda-
tionalism, 218
on the distinction between
logical and grammatical
form, 236, 449
on formal truth, 241
on paradoxes relating to
semantics, 280
on the meaning of “Hespe-
rus”, 298
on personal identity, 340–343
on the nature of propositions,
419, 443
on universals, 426
on contextual definition,
433–434, 448–449, 496
S
Salmon, Nathan:
on content-externalism, 4
on semantic externalism, 4
on the possibility of rationally
accepting self-contradicto-
ry propositions, 178
on Grice’s semantics, 417
Salmon, Wesley: 113
Sartre, Jean-Paul:
on the unconscious, 381–396
on repression, 390–396
See also Unconscious
thought.
Saussure, Ferdinand de: 312, 410
Schlick, Moritz:
on dynamic versus kinematic
relativity, 7
on event-individuation and
causality, 343

Searle, John:
on subpersonal thought, 31,
326, 335, 338–358
on conception, 115, 147–151
debate with McDowell,
147–151
on Grice’s semantics, 155–156,
483–484, 501
on the need to understand
symbols in psychological
terms, 261
on the Homunculus-objec-
tion to cognitive science,
338–342
debate with Chomsky
concerning subpersonal
thought, 338–358
on the difference between fol-
lowing and merely accord-
ing with rules, 338–358
on the Symbolic Conception
of Thought, 409
on the mind-dependent na-
ture of social facts, 486
positive view as to what
semantic rules are, 487
on the difference between
constitutive and regulatory
rules, 488
Sellars, Wilfred:
on the relationship between
phenomenology and
representational content,
31–32
on epistemological founda-
tionalism, 217
on inferential role semantics,
331
Semantic externalism: 3–214
as different from content-
externalism, 4
definition of, 4–5
Semantic rules: 18–19, 27–108
See also Pre-semantic impli-
cature
Sense-perception: 25–40
See also Non-conceptual
content
Sentences.
See Analogue-digital Distinc-
tion.
Shoemaker, Sydney: 339
Shope, Robert: 293
Sklar, Lawrence: 113
Soames, Scott:
on content-externalism, 4
on semantic externalism, 4
on whether semantic rules
must be known or oth-
erwise psychologically
represented, 27, 389
on semantic descriptivism,
41, 107
on wide-scope descriptiv-
ism, 53
on Frege’s paradox in relation
to the concept of direct-
reference, 137
Stich, Stephen:
on content-externalism and
its relation to folk-psychol-
ogy, 5, 23, 136
on conceptual non-atomism,
217
on functionalism, 258
Strawson, Galen: 213
Strawson, Peter:
on the Theory of Descrip-
tions, 76
on descriptivist theories of
perception in relation to
reduplication-problems,
149
on the gap between logical
and syntactical form,
236–237
on personal identity, 339
Stroud, Barry: 25
Subpersonal knowledge: 31, 76,
316, 323–326, 335–373
See also Unconscious
thought.
Substitution-failures: 71–95
Russell’s views on, 73–84
Frege’s views on, 84–95
See also Boguslawski,
Andrzej.
Symbols: 261–284
as partly psychological enti-
ties, 261
as partly non-psychological
entities, 261–262
likened to money, 263–265
the morphologically elastic
nature of, 265–275
likened to money, 263–265
Index 
as relations between psycho-
logical, sociological, and
morphological states of
affairs, 263, 273
definition of, 273, 505–506
in relation to the Compu-
tational Theory of Mind,
273–285
in relation to the possibility
of syntactic characteriza-
tions of semantic notions,
272–275
in relation to the regress-
argument against the
Symbolic Conception of
Thought, 280–284
See also Symbolic Concep-
tion of Thought.
Symbolic Conception of
Thought:
in relation to the Compu-
tational Theory of Mind,
9–10
regress-argument against,
280–284
arguments for, 391–402
arguments against, 273–285,
409–466
in relation to Fodor’s syn-
tactic analysis of thought,
278–279
as inconsistent with what
languages are, 438–440
as inconsistent with what
propositions are, 463
See also Symbols.
Synonymy: 79–82
Syntax:
as meaning-how, 13–15,
237–238
as combinatorial semantics,
14
as defined by mathematical
logicians, 14, 241–242
as defined by linguists, 14–15,
239–240
in relation to indexicality,
15–16, 237
in relation to morphology,
16–17, 233–235
as understood by advocates of
the Computational Theory
of Mind, 233–235
 Conceptual Atomism and the Computational Theory of Mind
in relation to semantics in
general, 235
as semantic decomposition,
239, 501
in relation to semantics, 235
See also Fodor, Jerry.
Systematicity of language and
thought: 397–401, 427
T work, 335–380
Taylor, Kenneth: 441, 453
Type-semantics:
in relation to indexicals, 69
as coinciding with pre-
semantics, 70–71
See also Kaplan, David.
Token-semantics: 69–70
See also Kaplan, David.
Transitional role semantics:
333–334
Truth:
as the instantiatedness of
properties, 454–456
why semantic rules oughtn’t
be understood in terms of
the concept of, 485–487
Truth-maker: 18–20
See also Content-externalism.
Tughendat, Ernst: 491, 497
U compute, 228
Unconscious thought: 335–396
the subpersonal unconscious,
335–380
in relation to Chomsky’s
the personal unconscious,
381–396
in relation to Freud’s work,
380–396
W
Wittgenstein, Ludwig:
in relation to Gricean theo-
ries of meaning, 156
in relation to the possibility
of referring to internal
states, 161
on functionalism, 212
on attitudinal eliminitavism,
213
on conceptual non-atomism,
217–218, 345
on epistemological founda-
tionalism, 217–218
on why computers don’t
on the conventional nature of
the relationship between
word and word-meaning,
228, 495
on mechanical procedures,
228, 249–250
on the definability of psycho-
logical terms, 258–259
on private language, 354–359
on rule-following, 361–367
on the subpersonal uncon-
scious, 326, 335, 357–373
as advocate of the view that
concepts are necessarily
unfrozen, 351
on Freud and the personal
unconscious, 373
on whether non-lexical items
refer, 492
Advances in Consciousness Research
A complete list of titles in this series can be found on the publishers’ website, www.benjamins.com

71 Krois, John Michael, Mats Rosengren, Angela Steidele and Dirk Westerkamp (eds.):
Embodiment in Cognition and Culture. 2007. xxii, 304 pp.
70 Rakover, Sam S.: To Understand a Cat. Methodology and philosophy. xviii, 246 pp. + index. Expected
August 2007
69 Kuczynski, John-Michael: Conceptual Atomism and the Computational Theory of Mind. A defense of
content-internalism and semantic externalism. x, 524 pp. Expected August 2007
68 Bråten, Stein (ed.): On Being Moved. From mirror neurons to empathy. 2007. x, 333 pp.
67 Albertazzi, Liliana (ed.): Visual Thought. The depictive space of perception. 2006. xii, 380 pp.
66 Vecchi, Tomaso and Gabriella Bottini (eds.): Imagery and Spatial Cognition. Methods, models and
cognitive assessment. 2006. xiv, 436 pp.
65 Shaumyan, Sebastian: Signs, Mind, and Reality. A theory of language as the folk model of the world.
2006. xxvii, 315 pp.
64 Hurlburt, Russell T. and Christopher L. Heavey: Exploring Inner Experience. The descriptive
experience sampling method. 2006. xii, 276 pp.
63 Bartsch, Renate: Memory and Understanding. Concept formation in Proust’s A la recherche du temps
perdu. 2005. x, 160 pp.
62 De Preester, Helena and Veroniek Knockaert (eds.): Body Image and Body Schema.
Interdisciplinary perspectives on the body. 2005. x, 346 pp.
61 Ellis, Ralph D.: Curious Emotions. Roots of consciousness and personality in motivated action. 2005.
viii, 240 pp.
60 Dietrich, Eric and Valerie Gray Hardcastle: Sisyphus’s Boulder. Consciousness and the limits of
the knowable. 2005. xii, 136 pp.
59 Zahavi, Dan, Thor Grünbaum and Josef Parnas (eds.): The Structure and Development of Self-
Consciousness. Interdisciplinary perspectives. 2004. xiv, 162 pp.
58 Globus, Gordon G., Karl H. Pribram and Giuseppe Vitiello (eds.): Brain and Being. At the
boundary between science, philosophy, language and arts. 2004. xii, 350 pp.
57 Wildgen, Wolfgang: The Evolution of Human Language. Scenarios, principles, and cultural dynamics.
2004. xii, 240 pp.
56 Gennaro, Rocco J. (ed.): Higher-Order Theories of Consciousness. An Anthology. 2004. xii, 371 pp.
55 Peruzzi, Alberto (ed.): Mind and Causality. 2004. xiv, 235 pp.
54 Beauregard, Mario (ed.): Consciousness, Emotional Self-Regulation and the Brain. 2004. xii, 294 pp.
53 Hatwell, Yvette, Arlette Streri and Edouard Gentaz (eds.): Touching for Knowing. Cognitive
psychology of haptic manual perception. 2003. x, 322 pp.
52 Northoff, Georg: Philosophy of the Brain. The brain problem. 2004. x, 433 pp.
51 Droege, Paula: Caging the Beast. A theory of sensory consciousness. 2003. x, 183 pp.
50 Globus, Gordon G.: Quantum Closures and Disclosures. Thinking-together postphenomenology and
quantum brain dynamics. 2003. xxii, 200 pp.
49 Osaka, Naoyuki (ed.): Neural Basis of Consciousness. 2003. viii, 227 pp.
48 Jiménez, Luis (ed.): Attention and Implicit Learning. 2003. x, 385 pp.
47 Cook, Norman D.: Tone of Voice and Mind. The connections between intonation, emotion, cognition and
consciousness. 2002. x, 293 pp.
46 Mateas, Michael and Phoebe Sengers (eds.): Narrative Intelligence. 2003. viii, 342 pp.
45 Dokic, Jérôme and Joëlle Proust (eds.): Simulation and Knowledge of Action. 2002. xxii, 271 pp.
44 Moore, Simon C. and Mike Oaksford (eds.): Emotional Cognition. From brain to behaviour. 2002.
vi, 350 pp.
43 Depraz, Nathalie, Francisco J. Varela and Pierre Vermersch: On Becoming Aware. A pragmatics
of experiencing. 2003. viii, 283 pp.
42 Stamenov, Maxim I. and Vittorio Gallese (eds.): Mirror Neurons and the Evolution of Brain and
Language. 2002. viii, 392 pp.
41 Albertazzi, Liliana (ed.): Unfolding Perceptual Continua. 2002. vi, 296 pp.
40 Mandler, George: Consciousness Recovered. Psychological functions and origins of conscious thought.
2002. xii, 142 pp.
39 Bartsch, Renate: Consciousness Emerging. The dynamics of perception, imagination, action, memory,
thought, and language. 2002. x, 258 pp.
38 Salzarulo, Piero and Gianluca Ficca (eds.): Awakening and Sleep–Wake Cycle Across Development.
2002. vi, 283 pp.
37 Pylkkänen, Paavo and Tere Vadén (eds.): Dimensions of Conscious Experience. 2001. xiv, 209 pp.
36 Perry, Elaine, Heather Ashton and Allan H. Young (eds.): Neurochemistry of Consciousness.
Neurotransmitters in mind. With a foreword by Susan Greenfield. 2002. xii, 344 pp.
35 Mc Kevitt, Paul, Seán Ó Nualláin and Conn Mulvihill (eds.): Language, Vision and Music.
Selected papers from the 8th International Workshop on the Cognitive Science of Natural Language
Processing, Galway, 1999. 2002. xii, 433 pp.
34 Fetzer, James H. (ed.): Consciousness Evolving. 2002. xx, 253 pp.
33 Yasue, Kunio, Mari Jibu and Tarcisio Della Senta (eds.): No Matter, Never Mind. Proceedings of
Toward a Science of Consciousness: Fundamental approaches, Tokyo 1999. 2002. xvi, 391 pp.
32 Vitiello, Giuseppe: My Double Unveiled. The dissipative quantum model of brain. 2001. xvi, 163 pp.
31 Rakover, Sam S. and Baruch Cahlon: Face Recognition. Cognitive and computational processes.
2001. x, 306 pp.
30 Brook, Andrew and Richard C. DeVidi (eds.): Self-Reference and Self-Awareness. 2001. viii, 277 pp.
29 Van Loocke, Philip (ed.): The Physical Nature of Consciousness. 2001. viii, 321 pp.
28 Zachar, Peter: Psychological Concepts and Biological Psychiatry. A philosophical analysis. 2000.
xx, 342 pp.
27 Gillett, Grant R. and John McMillan: Consciousness and Intentionality. 2001. x, 265 pp.
26 Ó Nualláin, Seán (ed.): Spatial Cognition. Foundations and applications. 2000. xvi, 366 pp.
25 Bachmann, Talis: Microgenetic Approach to the Conscious Mind. 2000. xiv, 300 pp.
24 Rovee-Collier, Carolyn, Harlene Hayne and Michael Colombo: The Development of Implicit
and Explicit Memory. 2000. x, 324 pp.
23 Zahavi, Dan (ed.): Exploring the Self. Philosophical and psychopathological perspectives on self-
experience. 2000. viii, 301 pp.
22 Rossetti, Yves and Antti Revonsuo (eds.): Beyond Dissociation. Interaction between dissociated
implicit and explicit processing. 2000. x, 372 pp.
21 Hutto, Daniel D.: Beyond Physicalism. 2000. xvi, 306 pp.
20 Kunzendorf, Robert G. and Benjamin Wallace (eds.): Individual Differences in Conscious
Experience. 2000. xii, 412 pp.
19 Dautenhahn, Kerstin (ed.): Human Cognition and Social Agent Technology. 2000. xxiv, 448 pp.
18 Palmer, Gary B. and Debra J. Occhi (eds.): Languages of Sentiment. Cultural constructions of
emotional substrates. 1999. vi, 272 pp.
17 Hutto, Daniel D.: The Presence of Mind. 1999. xiv, 252 pp.
16 Ellis, Ralph D. and Natika Newton (eds.): The Caldron of Consciousness. Motivation, affect and self-
organization — An anthology. 2000. xxii, 276 pp.
15 Challis, Bradford H. and Boris M. Velichkovsky (eds.): Stratification in Cognition and
Consciousness. 1999. viii, 293 pp.
14 Sheets-Johnstone, Maxine: The Primacy of Movement. 1999. xxxiv, 583 pp.
13 Velmans, Max (ed.): Investigating Phenomenal Consciousness. New methodologies and maps. 2000.
xii, 381 pp.
12 Stamenov, Maxim I. (ed.): Language Structure, Discourse and the Access to Consciousness. 1997.
xii, 364 pp.
11 Pylkkö, Pauli: The Aconceptual Mind. Heideggerian themes in holistic naturalism. 1998. xxvi, 297 pp.
10 Newton, Natika: Foundations of Understanding. 1996. x, 211 pp.
9 Ó Nualláin, Seán, Paul Mc Kevitt and Eoghan Mac Aogáin (eds.): Two Sciences of Mind.
Readings in cognitive science and consciousness. 1997. xii, 490 pp.
8 Grossenbacher, Peter G. (ed.): Finding Consciousness in the Brain. A neurocognitive approach. 2001.
xvi, 326 pp.
7 Mac Cormac, Earl and Maxim I. Stamenov (eds.): Fractals of Brain, Fractals of Mind. In search of a
symmetry bond. 1996. x, 359 pp.
6 Gennaro, Rocco J.: Consciousness and Self-Consciousness. A defense of the higher-order thought theory
of consciousness. 1996. x, 220 pp.
5 Stubenberg, Leopold: Consciousness and Qualia. 1998. x, 368 pp.
4 Hardcastle, Valerie Gray: Locating Consciousness. 1995. xviii, 266 pp.
3 Jibu, Mari and Kunio Yasue: Quantum Brain Dynamics and Consciousness. An introduction. 1995.
xvi, 244 pp.
2 Ellis, Ralph D.: Questioning Consciousness. The interplay of imagery, cognition, and emotion in the
human brain. 1995. viii, 262 pp.
1 Globus, Gordon G.: The Postmodern Brain. 1995. xii, 188 pp.